Abstract
This paper examines how equity claims were mobilized in the public debate over New York City’s Central Business District Tolling Program (CBDTP). We analyze approximately 8300 public comments submitted during the environmental review and use large language models (LLMs) for stance/equity classification followed by structural topic modeling to surface latent themes. Proponents’ equity arguments clustered around a compact set of frames: drivers paying their fair share, non-drivers reclaiming public space and road safety, air-quality improvements, and benefits to transit riders. Opponents advanced more varied arguments, ranging from fee regressivity and affordability pressures to distrust in the Metropolitan Transportation Authority’s management. Many opponents framed the fee as inequitable where driving is not discretionary—for example, medical/disability trips, off-peak and emergency workers with equipment, and contexts with unsafe or unreliable transit. These objections reflect Sandel’s critique of the market society, where willingness to pay (WTP) is mistaken for value, while ability to pay is overlooked. We extend this insight to transportation by highlighting unequal abilities to avoid paying. Even in New York, many commenters characterized the toll as an unavoidable levy. Congestion pricing’s legitimacy may depend not just on distributional effects, but also on whether payment is experienced as a genuine choice. Public comments reveal the enduring relevance of two choice-enhancing strategies in the literature: funding new mobility options for those who cannot pay, or preserving non-tolled alternatives for those who would rather wait than pay. If these concerns are being raised in New York, they are likely to dominate debate elsewhere in North America.
Introduction
Cities in Asia and Europe have implemented congestion pricing to promote sustainable travel, improve air quality, and fund transit service (Hess and Börjesson, 2019). North American policymakers have long advocated for the approach without success until 2025, when New York City implemented the continent’s first system. The impact of a new financial charge on commuters has motivated research on its distributional impacts and how unequal outcomes might be ameliorated. In the auto-dependent United States, researchers argue that the revenue generated from congestion pricing schemes must help offset the financial burden on households with low incomes (King et al., 2007; Levinson, 2010; Manville and Goldman, 2018; Small, 1992), though some have characterized this solution as impractical (Poole, 2011).
New York’s system, the Central Business District Tolling Program or CBDTP, charges vehicles for entering any part of Manhattan below 60th Street. For all vehicles with E-ZPASS toll transponders, the program charges $9 between 5am and 9pm during weekdays, and $2.25 at other times, with discounts for low-income New Yorkers, people with disabilities, and emergency vehicles. During the environmental review, however, prices of $15 and $22 were under consideration. Before its implementation, polling showed consistently high initial opposition statewide, including majorities in New York City (NYC) as defined by pollsters (Sienna College Research Institute, 2024a: 5, 2024b: 5). New York Governor Kathleen Hochul put the program on “indefinite pause” shortly before the 2024 election, only to implement it—with lower fees—in January 2025. After this change, Sienna College estimated that 56% of NYC residents still opposed the program just weeks prior to implementation (Sienna College Research Institute, 2024c: 6). Since that time, the program’s success in reducing emissions, traffic, and traffic violence (The Office of Governor Kathy Hochul, 2025) may be shifting public opinion in favor of the program (Marcelo, 2025).
The initial political backlash included protests by erstwhile progressive groups and members of the public concerned about the fairness of the scheme, including a lawsuit filed by the United Federation of Teachers (Matthews, 2024). Supporters of the program also invoked equity and fairness in making the public case for the program (Banks and Norton, 2025). Both sides’ claims to be advancing equity raise questions as to how the public understood the equity implications of the program and why this understanding varied so much. To unpack this puzzle, this study answers the following questions: How did participants in the public debate over congestion pricing make equity-related claims, what aspects of the program or its context informed those claims, and how do the justice frameworks reflected in their arguments align with or challenge existing theories in transportation and beyond?
This study analyzes over 8000 public comments submitted to the New York Metropolitan Transportation Authority (MTA) during the agency’s environmental review of the program (March 2021–June 2023). We first deploy large language models (LLMs) to classify each comment with respect to its position on the program and whether the comment raises an equity or fairness related point. We then adopt structural topic modeling, a form of “unsupervised machine learning,” to identify latent topics across public comments, which we cross-tabulate with our other codes to uncover why different commenters supported or opposed the program on equity or fairness grounds.
To provide context, we begin with an overview of the concept of congestion pricing and a synthesis of studies on its distributional equity impacts, focusing on the impacts included and assumptions made. Next, we introduce a critique of the expansion of markets into previously non-marketized aspects of daily life articulated by Sandel (2012), focusing on his discussion of queue jumping. Finally, we trace the CBDTP program’s evolution through the environmental review process, contextualizing the subsequent data, methods, and results sections.
Literature
Congestion pricing applies economic theory to the external costs of driving. These schemes aim to make drivers pay for marginal costs that their driving imposes on other drivers (i.e., by slowing other drivers down), as well as society (i.e., noise pollution; Button, 2004; Small and Gómez-Ibá ñez, 2005). The pricing system proposed for New York City represents a version of “downtown congestion pricing” (DCP), or programs that charge tolls for vehicles entering central business districts (Lehe, 2019). In a review of five systems, including two referenced frequently in debates in New York City, Lehe (2019: 201) finds that (1) the introduction of tolls coincided with investments in non-auto modes, and (2) that their political feasibility emerged from uncommon political events (e.g., the formation of a new local government authority). Lehe concludes that gradual efforts to secure political support do not lead to DCP implementation. A subsequent review of both successful and failed DCP proposals explains why. Selmoune et al. (2020) identify four factors driving public acceptance: personal privacy concerns, perceived fairness, uncertainty around promised benefits, and the complexity of implementation. The present study examines public sentiment through the lens of the second factor, perceived fairness. We turn next to review how these distributional impacts are tallied before introducing a critique of market-based policies from political philosophy.
Bounded by rationality: Equity analyses of congestion pricing
Most studies on the equity impacts of congestion pricing draw from distributional theories of justice to translate modeling results into policy-relevant findings. Before introducing critiques, we summarize this literature. Distributional justice in transportation examines how the impacts of transportation investments are allocated across the population (Pereira and Karner, 2021). Most studies examine distributional impacts through a regressivity/progressivity lens. This framing originates in tax policy, wherein a tax is regressive if lower-income households than pay a greater share of their income on the tax compared to higher-income households. Many transportation researchers and economists apply regressivity/progressivity framings more broadly, including when evaluating the distributional effects of road user fees (Foster, 2025), gas taxes (Glaeser et al., 2023), parking permits (Groote et al., 2016), and bridge tolls (Franklin, 2006). However, some have argued regressivity/progressivity is inappropriate in these contexts because these are fees for services or penalties rather than something everyone is subject to, like a tax (Manville, 2014). For this paper, we will use the language of regressivity because it is ubiquitous in the congestion pricing literature.
Distributional studies of European DCPs confirm the regressive nature of the charges (Bureau and Glachant, 2008; Di Ciommo and Lucas, 2014; Eliasson, 2016; Karlström and Franklin, 2009; West and Börjesson, 2020), but the literature offers several noteworthy exceptions. First, Stockholm’s system appears to be progressive because higher-income residents drive more into the CBD while lower-income residents benefit from spending on public transit (Eliasson and Mattsson, 2006; Franklin et al., 2016). Second, in London, researchers deploying ex-post evaluation found slightly progressive impacts because higher-income drivers comprised the majority of charged drivers (Craik and Balakrishnan, 2023). The study’s authors also noted that low-income drivers traveled to Central London much less than higher-income households after charges were later increased (Craik and Balakrishnan, 2023: 1027). Thus static equity analyses may not rigorously capture impacts over time. Research outside of Europe offers hope for more progressive outcomes. In China, modeling efforts suggest congestion programs can be progressive (Linn et al., 2016).
US-based research focuses mostly on toll roads and tolled lanes, which often ignite debates over the fairness of road pricing (Weinstein and Sciara, 2006). Schweitzer and Taylor (2008) argue that while congestion tolls are regressive when viewed in isolation, they are less inequitable than sales taxes, another major tool US localities use to fund transportation. Unlike road pricing, sales taxes require low-income non-drivers to subsidize frequent road users; in this sense, the equity arguments against congestion pricing are weaker than often claimed. Further, Plotnick et al. (2011) show that while tolls appear highly regressive when measured only among facility users, their distributional impact is far less severe when assessed across all households in a region. The governance structure can also impact equity, as multimodal transportation authorities are better able to reinvest revenues in transit and equity programs compared to private operators constrained by debt covenants (Weinreich, 2021). However, not all US-based studies are as positive. An analysis of driver responses to toll roads in Hampton Roads, VA, found Black and Asian drivers were more likely to engage in toll-avoidance behaviors, highlighting possible negative impacts to businesses catering to these populations (Yusuf et al., 2022). In sum, US tolling literature suggests that the equity of road pricing is not inherent to the tool itself, but depends on its design, governance, populations affected, and the distributional effects of alternative funding sources.
Recent modeling of behavioral data from the US argues that congestion pricing may be progressive for that country. One study focusing on Los Angeles lays out three arguments for this position. First, auto drivers in poverty are few enough that offering them exemptions would not harm policy objectives. Second, peak-hour drivers are wealthier than off-peak drivers. Third, the overrepresentation of low-income households near major roads in the US means that the air quality benefits would accrue disproportionately to these households (Manville and Goldman, 2018). A later study drawing on multiple data sources estimates welfare effects from tolls to provide rare US-based evidence that congestion pricing need not harm disadvantaged drivers. With the right design—such as leaving untolled alternatives or reinvesting revenues—pricing can even yield Pareto improvements, meaning some people gain welfare, while nobody becomes worse off (Hall, 2021). No other equity studies included in this review accounted for public health impacts and travel time savings.
Incorporating broader impacts
Congestion pricing can also reduce other externalities of congestion in ways that impact the distribution of benefits and burdens beyond monetary and time costs. London’s congestion pricing scheme reduced air pollution in the toll zone (Atkinson et al., 2009), and this effect persisted for years (Conte Keivabu and Rüttenauer, 2022). Absences from class also declined in schools with more students who are economically disadvantaged (Conte Keivabu and Rüttenauer, 2022). The potential for travelers to shift to more active modes of travel has led some to argue congestion pricing may bring other public health benefits (Levy et al., 2010), but mostly weak evidence supports this claim (Brown et al., 2015). A recent study found that active travel rates increased in communities along the border of London’s charging zone, with the effect greater for individuals with lower incomes (Nakamura et al., 2024). A scoping review on the broader impacts of congestion pricing concludes that congestion pricing has either neutral or positive effects on outcomes ranging from life expectancy and asthma attacks to air pollution and levels of social interaction (Hosford et al., 2021). While distributional studies clarify who pays and who benefits, they are largely silent on whether the price mechanism operates as a genuine choice. Political philosophers have argued, however, that relying on pricing to improve welfare ignores that not everyone has the same ability to pay.
Sandel, the morality of markets, and queue jumping
Sandel’s arguments on the moral limits of markets are motivated by a concern that use of markets, and what he terms “market values,” are increasingly moving “into spheres of life where they don’t belong” (Sandel, 2012: 7). Further, he notes that “In a society where everything is for sale, life is harder for those of modest means” (Sandel, 2012: 8). As life becomes more marketized, money becomes more important. As things become commodified, our attitudes toward them may change. For example, paying children to read may lead them to view reading as a chore. Finally, as markets become ubiquitous, their values crowd out other values a society may care about.
Sandel explores the implications of toll lanes, fast-track line passes at theme parks, and ticket scalping, collectively as what he calls marketized queue jumping. At its core, his concern with the marketization of queues rests on the economists’ assumption that willingness to pay is the best way of determining who most values a good. He argues that prices reflect both the willingness and the ability to pay (Sandel, 2012: 31). He notes one could just as easily argue that the best measure of who really desires to attend a concert is who is willing to stand in line the longest. Admittedly, this would privilege those with the most time, “but only in the same sense that markets ‘discriminate’ in favor of people who have the most money” (Sandel, 2012: 32). As we will show, this tension helps explain why both proponents and opponents of congestion pricing invoked equity in their arguments. The New York case provides a vivid illustration of how these abstract concerns about markets and fairness surfaced in a real policy debate.
Case study
New York City’s congestion pricing scheme charges vehicles for entering Manhattan below 60th street, several blocks below Central Park. The system uses an existing electronic transponder system, EZPASS, to charge drivers $9 every time they entered the congestion zone between 5 am and 9 pm, and $2.75 at any other time of day. App-based for hire vehicles pay $1.50 and taxis pay $ 0.75. Low-income vehicle owners receive a 50% discount for all trips made after the first 10 trips completed in the calendar month, with the discount applying for the rest of the calendar month thereafter. Individuals must make less than $50,000 per year or demonstrate enrollment in a qualifying government assistance program to participate. The program helps fund a Capital Improvement Program (CIP) that will fund improving accessibility for people with disabilities, improvements to subway cars, the system’s power grid, signals, and tracks, and construction of a new line, the Second Avenue Subway line (MTA, 2023). Once legislation was passed, they had to conduct an environmental assessment (EA) that required public feedback. This feedback forms the basis of our data.
Methods
We use public comments on the CBDTP Environmental Assessment (EA) to understand the public’s attitudes toward the proposal. Under the US National Environmental Policy Act (NEPA), agencies conduct an EA (40 CFR §1506.1(h)) to determine the significance of the project’s environmental impacts. If the assessment finds the impacts are not significant or that they can mitigate the adverse effects to a level of insignificance, then the FHWA, the lead federal agency for this project, issues a Finding of No Significant Impact (FONSI) statement justifying the decision (40 CFR §1500.1(a)). Appendix 18C of the congestion scheme’s FONSI details every comment submitted to the MTA during the EA process. This study analyzes Appendix 18C comments to assess public attitudes toward congestion pricing. Our analytical approach nests multiple methods into a multi-step process outlined below.
Cleaning and coding comments
Generating analyzable data required a few preliminary steps. First, we converted the PDF of the appendix into a spreadsheet of comments using the pdftools package in R. This yielded 13,000 comments. Second, we identified and removed all comments with at least 20% text identical to another comment, shrinking the corpus to 11,979 comments. We made this decision after comparing the resulting corpus with thresholds set at 10%, 15%, 20%, and 25%. A 20% threshold balanced the removal of duplicate form letters with the retention of substantively distinct comments; reductions beyond this point produced minimal change in corpus size, suggesting diminishing returns.
Our primary research question is how and why commenters on both sides mobilized equity arguments; specifically, what equity claims they made, what program/context features they tied them to, and how those frames map to justice theories. To enable such an analysis, we classified each remaining comment as being supportive, opposed to, or mixed/unclear in its position on congestion pricing. Because of the size of the corpus, we used two LLMs, ChatGPT 4.0 and Claude, to complete this task, as LLMs are effective at classification-based qualitative coding (Bijker et al., 2024). Each LLM received the same prompt: “Code the following comment about congestion pricing in New York as ‘S’ for supports, ‘O,’ for opposes and ‘M’ for mixed or unclear. Only return the code letter (i.e., an S, O, or M).” The lead author also coded 300 comments at random as a human standard to benchmark the LLMs against, with Claude and ChatGPT providing an 89% and 88% agreement, respectively. We then compared each LLM’s results against each other, finding an 86.01% agreement rate. In 80% of the disagreements, ChatGPT perceived a comment as opposed, while Claude perceived it as mixed or unclear. Upon reviewing 300 of these comments, the lead author sided with Claude and broke these ties in Claude’s favor. The lead author also broke the remaining 318 ties manually.
As we were interested in equity-based arguments made during the debate, we then reduced the corpus to only comments speaking to equity, fairness, or justice. We then had Claude and ChatGPT code each of the remaining comments using this prompt: “This is a public comment on congestion pricing in New York City. If this comment speaks to or references issues of fairness, equity, or justice, return the value 1. If the comment does not speak to or reference issues of fairness, justice, or equity, return a 0.” We did not offer any further context on what equity, fairness, or justice might mean to avoid introducing our own bias into the coding. The two LLMs achieved a 90.86% agreement rate, with ChatGPT identifying 80.57% of comments as relating to equity, fairness, or justice compared to 75.22% according to Claude. Ties were broken by the lead author. LLM coding and agreement statistics are presented in the Online Supplemental Material.
We further reviewed how each LLM scored 300 random comments on equity concerns, as a quality-control. We found both models interpreted the prompt broadly, sometimes flagging comments that did not explicitly invoke equity or fairness but raised concerns with clear distributive or justice implications. To reduce ambiguity in the coding process, comments identified by both LLMs as related to equity, fairness, or justice were kept in the corpus. This represented 73.26% of all submitted comments, or 8298 submissions.
Structural topic modeling
We examined the content of equity-related comments using structural topic modeling via the stm package in R. Structural topic modeling (STM) has gained the attention of transportation researchers because of its ability to identify latent topics or themes across a range of different documents (Kuhn, 2018). For this reason, transportation researchers have found this approach useful in identifying common themes across open-ended survey comments (Bardutz and Bigazzi, 2022). Thus, this approach is well suited to identifying common topics across public comments.
STM requires further data cleaning and setting an assumed number of topics, before running the algorithm. We deployed the processing command in stm to clean the corpus and adopted a data-driven approach to selecting the number of topics, in line with prior transportation studies using the method (Bai et al., 2021; Tamakloe and Park, 2023). We selected the number of topics by fitting STM models over a grid of K (number of topics) and maximizing a combined interpretability score (harmonic mean of normalized semantic coherence and exclusivity), with held-out likelihood as a tiebreaker. For candidate Ks we assessed stability by aligning topics across multiple random initializations using the Hungarian algorithm on top-word Jaccard (and β-cosine) similarities, and we retained the smallest K that achieved high coherence–exclusivity balance, good held-out performance, and stable, interpretable topics. Statistics comparing performance at different levels of K are available in the Online Supplemental Material. This process yielded an optimum of 20 topics.
We then ran a series of diagnostics (Table 1). The model shows a standard coherence–exclusivity trade-off (median exclusivity ≈ 9.69, IQR ≈ 0.31; median coherence ≈ −76.3, IQR ≈ 27.6), with topics reasonably distinct lexically while retaining within-topic co-occurrence. A residual check indicates overdispersion (dispersion ≈ 4.39, p < 0.001), a limitation of the analysis. However, this is not surprising given we are analyzing short, heterogeneous texts and a large corpus. Predictive performance across replication runs is stable (expected held-out mean ≈ −6.88, SD ≈ 0.014). Importantly, seed robustness is strong: topic prevalence structures align well across seeds (median θ-cosine ≈ 0.814, IQR ≈ 0.065), and lexical identities are moderately stable (median FREX Jaccard ≈ 0.535, IQR ≈ 0.020), with ∼53% of topics meeting a stricter ≥ 0.60 Jaccard threshold. Overall, diagnostics show a well-behaved model whose topical structure and salient word sets are reproducible across initializations, supporting the reliability of the substantive interpretations reported in the main text. Finally, pairwise topic correlations are low overall, with only two pairings yielding correlations over 0.2 and none going above 0.25. This suggests the model is capturing distinct themes with limited redundancy across topics. The full topic–topic correlation matrix is provided in the Online Supplemental Material.
Model diagnostics.
Results
We first report the 20 latent topics from the STM (Table 2). For each topic, we list two sets of words: “FREX” terms—the highest-scoring words under a weight that balances exclusivity and frequency (i.e., words distinctive to a topic but not vanishingly rare)—and “Prob” terms—the highest-probability words in that topic’s word distribution. Together they indicate both what makes a topic unique (FREX) and what makes it common (Prob). Topic 1 is a low-substance “catchall” that primarily captures translated/form-letter artifacts (e.g., forwarded emails) rather than policy content. We retain it deliberately: allowing the model to sequester these messages improves separation among the remaining, policy-relevant topics and avoids the stronger ex-ante researcher intervention of manually removing comments. Substantive interpretation, therefore, focuses on Topics 2–20, while Topic 1 functions as a sink for boilerplates that would otherwise blur topical boundaries.
Summary of topics from STM model.
Comments in favor of congestion pricing that pertained to equity focused on a narrower range of topics than opposition comments. To illustrate which topics motivated supporters and opponents on equity grounds, we produced a heat map of the distribution of topics among pro, con, and mixed comments (Figure 1). Heatmap cells show the log2 over-index of each topic within a stance—that is, log2 (share of topic within stance ÷ share of topic overall). Red cells mean the topic is over-represented among commenters with that stance, while blue indicates underrepresentation. White dots flag the three most over-indexed topics per column. Row labels list the topic ID, overall mean topic proportion μ, and the number of comments with θ ≥ 0.10 (n ≥ 0.10). For clarity, topics are displayed in order of overall prevalence (most to least common) rather than by numeric topic ID; thus, Topic 14 appears first. Pro-congestion pricing arguments are concentrated in a small set of topics (large positive deviations), whereas mixed and con comments largely mirror the overall topical mix with smaller deviations. This pattern suggests a relatively coherent pro coalition (e.g., emphasizing specific equity/benefit frames) versus a more diffuse set of concerns among opponents and those whose positions were not clear or mixed. To answer our question of understanding how and why supporters and opponents both believed their positions advanced equity, we organize results by analyzing the topics each group was more likely to speak to in public comment.
Over and under-representation of supporters and opponents among different topics.
Supporters
Comments with the highest theta scores for Topic 10, a pro-leaning topic on the benefits of congestion pricing, reveal concerns about modal equity and drivers paying their fair share. As one commenter noted: Driving a car is an unnecessary luxury for the vast majority of drivers, not a right, and it should be taxed as one since permitting private cars to use the streets makes them less safe and healthy and efficient for other more environmentally responsible users. (Comment 4274)
Similarly, a self-identified cyclist noted: As a frequent cyclist into Manhattan for work, the amount of car traffic on my way puts my life in danger on a daily basis… let the drivers pay for their massive negative externalities that they impose on the rest of us. (Comment 3622)
The emphasis on modal equity is reflected in proponents’ invoking of Topic 5, which focuses in part on bicycle and pedestrian infrastructure. Several comments alluded to spatial inequities in street design, describing road space for cars as “a disproportionate misuse of valuable public space” (Comment 3513) and claiming that “it’s only fair that cars should pay for the massive amount of space they take up in a city where space is precious” (Comment 3855).
Related to modal disparities, these commenters also repeatedly highlighted the dominance of transit use among New Yorkers as a majoritarian argument for the program: “only a very small percentage of New Yorkers and commuters use cars daily. It is time to prioritize transit riders and make our streets safer for pedestrians and cyclists as well” (Comment 1296). The dominance of transit as a commute mode led some to frame congestion pricing as a way to save an underfunded public good for the city: “Like the majority of New Yorkers, I do not own a car and take public transit. Congestion pricing will fund this essential public good and make our city more livable for the vast majority” (Comment 8243). Many of these commenters also invoked distributional equity and regressivity concerns by pointing out that low-income New Yorkers rely on transit, while people who drive into Manhattan are overwhelmingly wealthy: “The benefits of congestion pricing far outweigh downsides for drivers. Low-income and essential workers overwhelmingly ride public transit and drivers are on average wealthier than transit riders” (Comment 11,895). Although comments scoring high for Topic 10 touched on diverse ways of thinking about equity, including distributional justice, environmental justice, and concerns about externalities, a focus on correcting inequities between driving and all other modes unified these comments.
Topic 15 highlights the nuances of advancing equity through congestion pricing, with proponents over-represented in making these arguments. Comments with high theta scores on this topic focused mostly on advocating environmental mitigation to the Bronx while supporting the program overall. As one commenter asked the MTA to: Please select an approach that limits exemptions and traffic diversions, especially truck traffic along the Cross Bronx Expressway and Staten Island Expressway, and work with communities to mitigate any such impacts before congestion pricing is implemented. Thank you for advancing this critical policy to cut congestion. (Comment 112)
Several of these commenters leveraged environmental arguments in favor of congestion pricing to push for funding mitigations in the Bronx: “For congestion pricing to authentically deliver on its promise for environmental justice, it must deliver on overall traffic and emission reductions in the Bronx and other EJ communities” (Comment 2557). Requested interventions ranged from capping the Cross-Bronx Expressway to replacing power plants in the borough with cleaner alternatives.
Finally, equity-related comments from proponents were also overrepresented among Topic 17, which focused on procedural issues and exemptions for motorcoaches and other businesses like power companies. Comments with high theta scores for Topic 17 emphasized the demographics of motorcoach ridership as reasons to grant an exemption on equity grounds: “Intercity buses carry a disproportionate percentage of low-income and minority passengers. Greyhound’s passenger profile illustrates that point. Greyhound is a majority minority bus service provider” (Comment 1397). In other cases, groups that supported the goals of the program argued that offering their company exemptions would help the program meet its goals: Con Edison supports the Central Business Tolling Programs’ policy goals of reducing vehicular traffic within the Central Business District… Con Edison submits that applying the emergency vehicle exemption to Con Edison [emergency repair] vehicles is in the public interest and aligns with the policy goals of the Program. (Comment 6256)
However, these comments generally did not speak to the equity implications of the program itself.
In sum, supporters’ equity reasoning was coherent and bounded to a smaller set of topics. Rather than drawing on a wide range of justice frames, proponents converged on a focused narrative of modal fairness—that drivers underpay relative to the social costs they impose on others, and that congestion pricing can restore balance by funding cleaner, safer, and more efficient alternatives.
Opponents
Opponents were more likely to speak to the most common topics raised in the corpus. Many opponents found the program unfair given what they perceived to be an already too-high cost of living, with associated high taxes and existing fees (Topic 14). A comment that scored high on this topic synthesized these concerns succinctly, “It’s already so expensive to live in NJ/NY area. They shouldn’t increase prices, they should be looking for ways to decrease prices and give people a break, especially with all the inflation” (Comment 103). Opponents were also more likely to evoke Topic 19, also framing the charge as harming already struggling people, but with class-based appeals: this program, if passed, will be detrimental to the working class - still reeling from COVID. The working class are still trying to get on our feet, and this will knock out so many jobs and the costs will be passed into us, once again. I vote no! (Comment 5340)
For these opponents, the program’s perceived injustice stemmed in part from its introduction into a broader economic context of inflation, inequality, and a perceived “squeeze” of New York’s middle class. These comments reflect Sandel’s concern about the commodification of society. As he notes, “In a society where everything is for sale, life is harder for those of modest means. The more money can buy, the more affluence (or the lack of it) matters” (Sandel, 2012: 8).
Interestingly, the most morally loaded arguments came from Topic 2, which blended two of the major factors driving public opinion on congestion pricing: perceived fairness and uncertainty around promised benefits (Selmoune et al., 2020). These opponents did not believe the MTA could deliver and instead accused the agency of trying “to line the pockets of executives of the MTA and nothing more” (Comment 2096). Relatedly, opponents also invoked Topic 5, which incorporated bike/ped infrastructure, to argue that the program did not address what they believed were the true causes of congestion in the city: I love this scam. Start by making bike lanes to narrow streets, then make outdoor street plazas, Citi Bike spaces and then as if a gift from God, have outdoor dining in streets to block off more street space. Now you can tell the public the streets are congested … The only solution is get money from drivers. (Comment 1781)
Distributional concerns also motivated opponents, who were more likely to invoke Topic 4, characterized in part by use of the word “regressive,” and Topic 20, which focused on household budgets and the exemption for low-income drivers. One resident argued that: The CBD tolling is a regressive tax. All NYC residents earning under $60,000 per year should be exempt, not just those who live in the area. The rich should not be able to buy faster rides for their chauffeur driven limousines by pricing out working New Yorkers. (Comment 5222)
Opponents were also more likely to invoke Topics 3, 6, and 13 to argue that congestion pricing is unfair to those who have no choice but to drive. As one commenter asked: What about people who live with or have disabled family or friends, and driving them is the best and safest way to get around? This is a regular occurrence in my family, as my mother cannot walk and has multiple doctor appointments a week. (Comment 371)
A woman whose family sought specific healthcare options in Chinatown stated that: We would have to compromise our medical care and pick and choose which appointments we could go without. There’s a great deal of unfairness and injustice that we are forced to sacrifice our medical care so uber and taxis can get free rides and MTA can raise money. (Comment 4390)
In other cases, the lack of options came down to schedule misalignment: “This initiative would cripple the hospitality industry. It is a take on people who have to work odd hours with limited means of transportation. Mass transit is simply unreliable in very early or very late hours” (Comment 6507). These arguments relate to Sandel’s concern that market behavior reflects not just willingness to pay but ability to pay (Sandel, 2012), or in this case, ability to avoid paying. For these travelers, the toll functions less as a nudge and more as a levy on a necessity. Equity objections here are not merely distributive; they center on unequal abilities to adapt when conditions such as health, disability, work hours, and the safety/reliability of service, constrain agency.
Conclusions
This study asked how participants in New York’s congestion pricing debate made equity-related claims and what features of the program and context informed those claims. Using LLM-assisted coding and structural topic modeling, we find that supporters’ equity claims clustered around a small set of themes—drivers paying their share, the road as public space safety, and air-quality mitigation. By weaving air quality and road safety benefits into a coherent narrative around modal equity for pedestrians, cyclists, and transit users, congestion pricing advocates are ahead of an empirical literature only now beginning to consider the total distributional effects of congestion pricing (Hosford et al., 2021).
These supportive claims carry their own implications. For research, they highlight the need to fully quantify modal equity—not only the direct financial transfers between drivers and non-drivers, but also the underpayment of motorists relative to the social costs of air pollution, crashes, and other harms of automobility. Developing rigorous measures of these externalities, and testing how pricing corrects them, would provide empirical clarity around the narratives advanced by supporters. For practice, they suggest that city planners should foreground the co-benefits of pricing—cleaner air, safer streets, and fairer space allocation—as central to the policy’s equity rationale as New York advocates did, rather than treating them as secondary outcomes. Making these benefits legible and immediate may help shift debates from cost burdens to shared gains.
Opponents, in contrast, raised a more diffuse set of concerns. Two of the most prevalent topics in the corpus framed the tolls as unfair within a broader economic context of rising costs and a perceived “squeeze” on the middle class. Planners developing congestion pricing systems must recognize that macroeconomic pressures like inflation and income inequality amplify equity objections. Policymakers should pair pricing with visible, near-term benefits—such as transit reliability upgrades or targeted relief—that make the policy legible as a public good rather than another tax.
Most importantly, equity objections frequently centered on constrained choice: Topics 3, 6, and 13 show commenters describing medical trips, off-peak and emergency work, jobs that require a vehicle, and unsafe or unreliable transit as conditions that make driving feel non-discretionary. Extending Sandel’s reasoning, prices reflect not only willingness-to-pay, but ability to avoid paying; where feasible alternatives are absent, a toll risks functioning as a levy on necessity rather than a price on discretion.
These findings carry two implications. First, New York—arguably the least coercive US setting for congestion charging given its transit scale—still produced numerous claims of constrained agency. In more auto-dependent cities, similar objections are likely to be more prevalent and vehement. Second, the role of perceived fairness in public acceptance hinges not only on distributional considerations, as found previously (Selmoune et al., 2020), but also on the extent to which the fee is perceived as a real choice and not another obligation for some drivers.
Public discourse in New York invites reflection on what a “choice-restoring” congestion pricing design might look like for other US cities. Commenters’ objections bring us back to two pathways that planners and policymakers continue to debate. One emphasizes targeted relief and reinvestment—limited exemptions for necessity-bound travel and revenue recycling that enhances transit service—an agenda of progressive redistribution that takes various forms in the literature (Manville and Goldman, 2018; Small, 1992). The other, represented in Hall’s (2021) work, pursues equity through structural design: maintaining untolled lanes, so that travelers who would rather wait than pay keep a real alternative. New York’s experience shows that even when elements of the first strategy are implemented, perceptions of constrained agency may persist, lending renewed relevance to the second. The challenge for other North American cities is less about choosing a design template and more about understanding which forms of “choice” the public recognizes as genuine.
Revisiting Sandel’s challenge to market-based reasoning: is willingness to pay—or willingness to wait—the best measure of who most values participation, whether at a concert or on the road? A profession that defines congestion as a problem has already taken a normative position on that question. While this logic may affirm professional priorities, we should not be dismissive when those with more time than money protest, especially when more equitable alternatives may exist (Hall, 2021).
Study limitations include the non-representativeness of public comments, imperfect classification from LLM-assisted coding (89% and 88% agreement rates with the human coder), and topic-model abstraction. Future work should track how individuals’ perceptions of the program evolve after implementation. More in-depth qualitative studies of changed perceptions may help planners identify how to communicate the program’s benefits before its introduction.
Well-designed pricing can reduce congestion and improve health and safety. It will be most legitimate where it funds new choices or preserves non-tolled options for those who may otherwise feel forced to pay congestion fees due to a lack of feasible alternatives. Planners in other North American cities should expect these concerns and plan to address them head-on.
Supplemental Material
sj-docx-1-usj-10.1177_00420980261419975 – Supplemental material for Americans meet congestion pricing: Reframing the equity debate around CBD tolling for auto-dependent countries
Supplemental material, sj-docx-1-usj-10.1177_00420980261419975 for Americans meet congestion pricing: Reframing the equity debate around CBD tolling for auto-dependent countries by Matthew Palm and Alainna Thomas in Urban Studies
Footnotes
Acknowledgements
We would like to acknowledge Wani Pandey who provided preliminary coding and analytical support in the early stages of the project.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the University of North Carolina, Chapel Hill and the Southeastern Transportation Research, Innovation, Development and Education Center (STRIDE).
Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Supplemental material
Supplemental material for this article is available online.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
