Compensation and negotiation for housing

The construction of new homes is rife with concentrated costs. Development brings noise and congestion, potentially harming quality of life. New residents may consume more in public services than they provide in tax revenue, raising the tax burden of existing property owners (Hamilton Reference Hamilton1976). Biases against racial outgroups may cause current residents to be wary of new neighbors (Charles Reference Charles2006). These threats – as well as a decline in property values from increased supply – may lead homeowners to oppose new housing (Fischel Reference Fischel2001). Similarly, renters may oppose new market-rate housing because they believe it will attract demand to their neighborhoods, increasing local rents (Hankinson Reference Hankinson2018; Nall et al. Reference Nall, Elmendorf and Oklobdzija2024).

Even when in the minority, local housing opponents are often effective in blocking or down-sizing proposed developments. Low-turnout local elections and permit review processes with unrepresentative public comment reward the preferences of organized, wealthier homeowners who often want no new housing, only single-family housing, or housing located outside of their neighborhoods (Einstein et al. Reference Einstein, Palmer and Glick2019). Collectively, these political barriers to housing development threaten equity both locally and nationally. Limiting new housing not only raises rents (see, e.g., Been et al. Reference Been, Ellen and O’Regan2019, for a review), but also prices out those seeking to move to cities with high upward income mobility, exacerbating income inequality (Ganong and Shoag Reference Ganong and Shoag2017) and entrenching racial segregation (Trounstine Reference Trounstine2018).

Although permitting decisions may be formally controlled by appointed officials, these officials are responsive to public comment on individual development projects (Sahn Reference Sahn2023). Thus, the mass public – and the reaction of the public to concentrated costs of housing development – meaningfully influences policy change. While efforts to persuade respondents of housing’s collective benefits have shown limited effectiveness (Marble and Nall Reference Marble and Nall2021), the concentrated costs of housing may be directly countered through concentrated benefits.

Historically, such benefits were public in nature and designed to offset the infrastructure costs of new development. Known as exactions or linkage fees, the amount of compensation is formula-based, limiting the ability of the surrounding community to secure additional benefits using their political leverage (Been Reference Been2005). Over time, however, the conceptualization of infrastructure and externalities has expanded to include effects on human capital. Today, even formula-based exactions may include public amenities beyond road and sewer development (Kim Reference Kim2020). Likewise, city governments have institutionalized the role of community groups in vocalizing what they would like to see from new development. These institutions formalize the process of negotiation over these collective benefits, giving political power to neighbors to secure compensation from developers in many cities.

How concentrated benefits are institutionalized

In line with larger efforts to enhance community voice within public administration (Bingham et al. Reference Bingham, Nabatchi and O’Leary2005; Jakobsen et al. Reference Jakobsen, James, Moynihan and Nabatchi2019; Vigoda Reference Vigoda2002), local governments have worked to better integrate citizen input into the housing approval process. But the inclusion of community input varies both by the type of housing proposed and the structure of the approval process within the city.

First, the permitting of housing differs based on the two types of proposals: by-right and discretionary. By-right proposals are those currently allowed by zoning and thus their approval is largely administrative, insulating them from community input. In contrast, proposals that exceed the zoning code are subject to discretionary review via a legislative body that will solicit community input. Because of the strictness of contemporary zoning, new housing developments increasingly must go through this discretionary review (O’Neill et al. Reference O’Neill, Gualco-Nelson and Biber2019, Reference O’Neill, Gualco-Nelson and Biber2020).

Second, how community input is institutionalized varies across cities. At the more limited end of the spectrum, discretionary review may be confined to a public meeting in front of the city’s Planning Commission – an appointed board composed of professionals such as architects or lawyers. During these meetings, residents may use public statements to attempt to change the development’s design or secure community benefits from the developer. In this context, residents are operating as individuals and not negotiating with the developer as a unified group. Following these public statements and any concessions offered by the developer, the commission will vote on whether to approve the permit.

There is considerable debate over the power community members have in these meetings. On one hand, public input may be disregarded and the meeting’s occurrence used as a form of “tokenization” to create a sense of democratic legitimacy (Arnstein Reference Arnstein1969; Checkoway Reference Checkoway1981). On the other hand, research has found evidence that community input does affect decision-making (Dynes et al. Reference Dynes, Karpowitz and Monson2022; Sahn Reference Sahn2023). Likewise, commission members often refer to community support in their rationale for approval or denial of projects (Einstein et al. Reference Einstein, Palmer and Glick2019). Even beyond the direct effect of public comment, these meetings may serve as venues of coordination and agenda-setting for future political action (Adams Reference Adams2004).

At the other end of the community input spectrum, a coalition of community groups may negotiate a formal community benefits agreement (CBA) with the developer. Ranging from financial to physical, to behavioral goods, CBAs may include the provision of affordable housing units or the guarantee of a living wage for employees who are residents of the community (Wolf-Powers Reference Wolf-Powers2010). In exchange, community groups will pledge to support the development, typically through testimony at public hearings. Compared to standardized benefits like exactions, this direct negotiation between developers and community groups is theoretically more efficient for securing community benefits and maintaining an elastic housing supply (Foster and Warren Reference Foster and Warren2022). As a result, a well-negotiated and legally enforceable CBA can provide a community with valuable resources while helping developers win political support.

Traditionally, formal CBAs have been confined to large, mixed-used developments on the scale of multiple city blocks. The uniqueness of these projects limits their comparability to each other, as well as the generalizability of their negotiations. Instead, we focus on community benefits that result from more common, semi-formalized negotiations between community groups and developers. These negotiations are similar to CBAs in that the city government recognizes a group of residents as representatives of the affected community, thus providing the agreement with legitimacy. But unlike CBAs, these negotiations occur in tandem with developments of all sizes that require discretionary review, including most multifamily housing.

How common are these opportunities for semi-formalized negotiation? We reviewed the discretionary review processes of the 25 most populous American cities and examined whether the following conditions exist:

1. 1. A structure of geographically defined groups recognized by city government as representing a neighborhood.

2. 2. Developers are required to meet with these groups as part of the discretionary review process.

3. 3. These groups are asked to supply formal recommendations regarding approval of the project.

Table A-1 in the Appendix outlines our findings. To summarize, 12 of the 25 most populous cities recognize a geographically defined entity as representing community interests in these decisions. In 8 of these 12 cities, community consultation is formally required in discretionary review. Within these 12 cities, we expect neighborhoods to be able to better exert their political influence and negotiate for compensation relative to cities without any recognized entity representing the community. In the other 13 cities, community groups may struggle to coordinate their negotiation efforts and risk developers splintering the community by selecting only favorable groups to represent the community – e.g., the controversial Atlantic Yards CBA (Been Reference Been2010).

But even within these 12 cities with community negotiation institutions, some cities have more formal community input. For instance, Boston, MA is known for heavily relying on negotiated benefits unique to each development, rather than scheduled benefits based on a fixed formula (Kim Reference Kim2020).Footnote ¹ More broadly, the institutionalization of community input appears to correspond with the ideology of city voters, with more conservative southern cities lacking formal recognition for community organizations in the development review process.

To understand how communities secure benefits from developers, we analyze 421 agreements from Boston, signed between 2016 and 2021. Known locally as “cooperation agreements,” these packages range from large amounts of money for community groups to other investments in physical infrastructure. Not all benefits are assigned financial values. For example, a development may provide a community group with a room for monthly meetings but not provide an estimated value of that benefit. Of the 421 agreements we observed, 35% provided some amount of financial compensation.

Of agreements including financial values, the average package was $240,000 with a maximum of $5.35 million. To estimate benefits per capita, we define beneficiaries as residents within the typical development notification radius – where developers need to notify residents about public hearings regarding their proposal. This radius is often ≤300 ft around the property. Given Boston’s population density of ∼14,000 people per square mile, the average payout would be $1,680 per person. In total, $35.7 million in specified financial benefits were committed to communities through these agreements in the six-year period we observed. Approximately 37% of this amount went to parks and recreation, 21% to community-based centers and resources, 28% to streets and transportation, and the remaining 13% to individual non-profits.

Figure 1 shows the distribution of these agreements across Boston, overlaid on a map of neighborhoods’ median household income levels. The agreements are both common and geographically dispersed. They exist in the wealthier neighborhoods along the harbor, the majority single-family home neighborhoods in the southwest of the city, and in the lower-income communities in the middle and southeast of the city. The volume and distribution of these agreements suggest that most neighborhoods in Boston have experience with these negotiations.

Figure 1.

Distribution of cooperation agreements, Boston, MA (2016–2021).

These observational data indicate how common it is for development review processes to involve material compensation in exchange for political support. But if we want to understand how these agreements shape political support, examining only the finalized agreements is a form of selection on the dependent variable. Data on real-world community benefits for successful projects inherently lack information about unsuccessful projects and the (unsatisfactory) package of compensation that they might have involved. Thus examining only successful agreements cannot shed light on whether such compensation is critical in securing support for new housing. Instead, using an experimental approach and randomly varying the size and structure of proposed compensation packages enables us to avoid selection bias and identify the causal effect of compensation on local support for development.

Experimental design

To assess the causal effect of compensation, we use a “willingness-to-accept” survey experiment wherein we show residents of Boston hypothetical new buildings proposed within their self-reported neighborhood and describe the randomly varied bundles of compensation that are offered by developers in exchange for their support. We conducted our preregistered experiment on a sample of over 578 Boston residents recruited through a variety of methods from April 2021 to April 2022.

This experimental design leverages both a willingness-to-accept framework and the spatial dimension of neighborhoods to mimic the real-world concentrated costs and potential benefits from housing development.Footnote ² To do so, our survey asked for respondents’ approximate home locations, calculated the distance between the proposals and respondents’ locations, and displayed 3-dimensional renderings of housing proposals on actual nearby residential parcels in their neighborhoods. The survey allowed the respondent to either enter their address or to first enter their ZIP code, then zooming in to their neighborhood.Footnote ³ Respondents were then asked to indicate the intersection nearest to their home.

Next, respondents were shown five development proposals randomly sampled from a list of potential proposals within a roughly 1 mile radius from their home. We chose nearby proposals in this distance range because spatially-driven opposition in an urban environment declines rapidly beyond this distance (e.g., Hankinson Reference Hankinson2018). Likewise, councilmembers considering whether to approve a proposal may provide greater weight to input from to those living closer to the housing proposal. In other words, proposals within this radius are the ones where respondents would both have the most leverage to obstruct and would be the most likely to benefit from compensation.

The visual presentation of these proposals was designed to mimic how proposals might be encountered in respondents’ daily lives. Each proposal contained two images: the existing parcel viewed from the sidewalk captured via Google’s Street View and a rendering of the proposed development (see Figure 2). The rendering was based on a three-dimensional representation of the current structure captured from slightly above via Google Earth. To represent the proposed building, each rendering included a blue prism drawn over the existing building to display the physical size of the proposed development without providing any details of its exterior design. These two images were displayed alongside a map showing a blue icon – the respondent’s location – and an orange icon – the location of the proposal. Throughout the survey, the blue icon always remained visible, with the screen reorienting to show the location of each new proposal.

Figure 2.

Example prompt. (a) Self-reported location of respondents (b) Location of building proposals.

The proposed developments were sampled from real residential structures that exist in the City of Boston’s property database, ensuring that only realistic locations for development were shown. Each proposal was described as twice as tall as the current building and containing threefold as many units. The proposal’s was stated in text and displayed using a blue prism surrounding the existing structure. The number of units in the proposal was also displayed in text and was rounded up for buildings with odd numbers of units. This increase in density was substantial but not unrealistic for new residential development in Boston.

Experimentally, the survey randomly varied three features of each proposal. First, we varied the affordability of each proposal’s units, stating either: “Half of the units would be occupied by low-income housing voucher recipients” or “The units will be rented at whatever price the local market supports.” Of course, the effects of affordable vs. market-rate development reach beyond the price point of individual units, and might include effects on the racial and economic diversity of the neighborhood and nearby schools. But this bundled treatment mirrors how affordable housing is often described by developers. At the bottom of the proposal, respondents were asked their support for the new building using a 5-item Likert response scale ranging from “Strongly oppose” to “Strongly support.” We rescaled this measure to range from 0 (least support) to 1 (most support) and used it as our first outcome of interest.

Next, respondents were given information about the compensation offered by the developer. We randomly varied the compensation amount, ranging from $50 to $10,000. These amounts were chosen to cover the median compensation level required for winning support and to avoid obviously excessive bids (Kanninen Reference Kanninen1995).Footnote ⁴ The median compensation bid offered was $1,500, which is in line with the earlier estimate of $1,680 per capita we calculated from the existing Boston cooperation agreements. This compensation was presented either as a direct payment to the respondent or as an investment in local public goods, randomized at the individual-level but held constant across each of the five proposals viewed to minimize cognitive load.

The text read:

The size of the “neighborhood” and total amount of compensation to be paid out ($X * # of neighborhood residents) was not defined, allowing respondents to form their own mental image of their neighborhood.

We might reasonably assume that respondents would prefer cash benefits for their fungibility. However, research on the use of compensation for similar land uses finds that residents do not like the feeling of being bribed. For example, Frey et al. (Reference Frey, Oberholzer-Gee and Eichenberger1996) find that residents are less supportive of a nearby nuclear waste facility when offered a small amount of compensation rather than no compensation at all. They argue that this happens because the financial payment crowds out the “warm glow” effects of doing one’s civic duty by accepting the waste facility. Experimental evidence on the siting of wind energy has also found that residents prefer public goods provision rather than private payments, likely due to the chronic under-provision of local public goods in these communities (García et al. Reference García, Cherry, Kallbekken and Torvanger2016). Ultimately, theoretical evidence points in both directions.

Following this information about compensation, we then asked respondents our second outcome measure for whether they supported the proposal. Following best practices of contingent valuation experiments, we phrased this measure in the form of a referendum. Respondents were asked “How would you vote on this proposal?” and indicated their support on a binary scale.

To summarize, the randomized features of the proposal were the following:

• Affordability: 0% of units for low-income residents v. 50% of units for low-income residents. Randomized at the proposal level.

• Compensation ranging from $50 to $10,000. Randomized at the proposal level.

• Form of compensation: Direct payment vs. public goods investment. Randomized at the respondent level.

Data

Our survey was designed only for respondents in Boston, due to the tractability of creating customized renderings of developments to serve as experimental stimuli. Yet Boston is an appropriate choice for studying opinions about housing policy: the city has high housing costs and struggles with siting new housing (Glaeser and Ward Reference Glaeser and Ward2009) in a similar fashion to many other large cities. And like many large American cities, Boston residents are predominantly liberal and Democratic. So while our survey respondents likely mirror the population most relevant for studying responses to housing development in large cities, we caution against extrapolating our findings to what we might find if we were to conduct our experiment on a broad national population.

We gathered responses from Boston residents via three methods. Wave 1 of the survey was fielded in April 2021. To recruit the sample, we used a commercially available voter file. We defined the sampling frame of registered voters living in Boston with an email address provided in the voter file (57% of registered voters). We used stratified sampling, grouping voters by race, age, voter turnout in the 2018 general election, and registered political party. We oversampled young and minority voters using estimated response rates from a similar voter file-based survey (Wilcox-Archuleta Reference Wilcox-Archuleta2019) to improve representativeness. Targeting a sample of 1,000 respondents, we emailed 46,833 voters. Participants who completed the 10-minute survey received a $5 Amazon gift card sent to their email address. Wave 1 contained 288 respondents.Footnote ⁵

Wave 2 of the survey was recruited via snowball sampling of neighborhood associations and tenant groups in Fall of 2021. We emailed unique survey IDs to individuals in leadership positions within Boston neighborhood associations and groups and asked them to distribute the survey to their members. No compensation was offered for this survey. While snowball sampling is by no means appropriate or ideal for gathering representative survey samples (Erickson Reference Erickson1979), the respondents for this wave were not necessarily intended to represent the broader city population. Instead, the target universe of respondents for this survey wave were those people most likely to attend community meetings to express their support or opposition to new housing development – and therefore those people whose opinions likely hold a great deal of power in the housing policy process in cities. In line with this expectation, 76% of Wave 2 respondents reported attending a Boston political meeting or community forum in the past 12 months, meaning their voice is incredibly relevant to this political phenomenon. Wave 2 contained 216 respondents.

Wave 3 of the survey was recruited to help maximize our sample size and therefore statistical power for our experiment. We fielded this wave in February 2022 via the PureSpectrum survey platform by targeting respondents registered with Boston-based ZIP codes. To ensure data quality, respondents were first filtered based on self-reported residence in Boston then respondents had to indicate their address within the city using the approach described above. These requirements make us confident that all respondents are current residents of Boston. Wave 3 contained 300 respondents.

We combined responses to Waves 1, 2, and 3 for a total of 805 respondents, of which 589 respondents provided demographic information. The demographics of our combined survey sample match the population of Boston reasonably well, as we show in Appendix Table H2.Footnote ⁶ Figure 3 shows the spatial distribution of respondents and our experimental stimuli across the Boston area. These maps show that both our respondents and the proposed developments that they evaluated encompassed nearly all of the city’s residential geography.

Figure 3.

Geographic distributions of respondents and experimental stimuli.

To analyze the experiment, we followed our pre-analysis plan and regressed support for each housing proposal (using our two separate outcome measures) on the randomly varied attributes of each development: compensation amount, inclusion of affordable housing, and form of compensation. We also included an array of demographic covariates including homeownership, income, race/ethnicity, education, partisanship, gender, and age. We used Huber-White standard errors clustered at the respondent level to account for the multiple proposals evaluated by each respondent.Footnote ⁷

The effects of affordability

To examine respondents’ baseline attitudes towards new housing, we look at support for proposals measured before compensation was described using the following ordinary least squares (OLS) equations (1) and (2):

(1) $${\rm{suppor}}{{\rm{t}}_{\rm{i}}} = {\beta _{\rm{0}}} + {\beta _{\rm{1}}}{\rm{distanc}}{{\rm{e}}_{\rm{i}}} + {\beta _{\rm{2}}}{\rm{affordabilit}}{{\rm{y}}_{\rm{i}}} + \gamma {{\bf{X}}_{\rm{i}}}{\rm{ + }}{\varepsilon _{\rm{I}}}$$

The dependent variable of support is the Likert response operationalized as an interval variable with a unit scale from 0 (“Strongly oppose”) to 1 (“Strongly support”). For the randomly varied proposal attributes, we operationalize distance based on a kilometer increase away from the respondent’s house, and “Affordability” as a dummy variable indicating whether the proposal included half affordable units, rather than all market-rate units. X _i is a vector of individual-level covariates including homeownership status, income, race, education, ideology, gender, and age. Figure 4 displays the effects of the randomly varied attributes on support for the proposal, both from models with demographic controls included (filled triangles) and without controls (filled circles).

Figure 4.

Predictors of support for housing proposals without compensation. Lines indicate 95%- confidence intervals (thin lines) and 90%-confidence intervals (thick lines).

Corroborating recent research, we find that the distance between a respondent’s home and the proposed development influences their support. A 1 km increase in the distance of the proposal away from a respondent’s home increased support by 0.05 to 0.08 (p < .05) along the 0–1 scale towards the highest outcome category of “strongly support.” In addition, proposals including affordable units were 0.07 to 0.08 (p < .001) more popular among respondents.Footnote ⁸ The positive effect of affordable housing is unexpected, given that past empirical work has found that affordable housing generally has less support than market-rate housing, especially among homeowners. We explore and consider mechanisms behind this effect in the Discussion.

The effects of compensation

We next assess the effect of compensation and other experimental features of the proposed developments. To do so, we use our second outcome, which measured binary support for the proposal after the development’s compensation package was described to respondents. On average, 49% of respondents reported support for this outcome for proposals in our experiment. We adapt Equation 1 by using our second outcome variable and by adding the experimental conditions of the amount and form of the compensation (public v. private benefits). We log compensation in order to assess the effects as percentage change in compensation rather than a nominal change.

(2) $$\matrix{ {{\rm{suppor}}{{\rm{t}}_{\rm{i}}} = {\beta _0} + {\beta _1}{\rm{log}}{{\left( {{\rm{compensation}}} \right)}_{\rm{i}}} + {\beta _2}{\rm{for}}{{\rm{m}}_{\rm{i}}} + {\beta _3}{\rm{distanc}}{{\rm{e}}_{\rm{i}}}} \hfill \cr {\quad \quad \quad \quad \quad \quad \quad \quad + {\beta _4}{\rm{affordabilit}}{{\rm{y}}_{\rm{i}}} + \gamma {{\bf{X}}_{\rm{i}}}{\rm{ + }}{\varepsilon _{\rm{I}}}} \hfill \cr } $$

We display the results of these analyses in Figure 5. Compensation increased respondents’ support for proposed developments. The median amount of compensation offered in our experiment was $1,500 per person. Doubling that compensation – meaning an increase in compensation from $1,500 to $3,000 per person – increased support for the proposal by 2.6 percentage points (p < .001). Relative to 49% average support for proposals, this type of effect is substantively significant given that it could help achieve majority support.

Figure 5.

Predictors of support for housing proposals with compensation. Lines indicate 95%-confidence intervals (thin lines) and 90%-confidence intervals (thick lines).

Whether that compensation was offered via a public benefit rather than a private payment appears to have had no detectable effect on respondents’ support, however. Providing the benefits as public goods rather than private payments had a positive effect in all models, but these effects were not statistically significant.Footnote ⁹ As in our earlier analyses, proposals that included affordable housing received more support from respondents. Respondents reported 9 percentage points higher support for partially affordable proposals than for market-rate proposals (p < .001).

However, the form of housing (affordable rather than market-rate) moderated the effect of compensation on respondents’ support for developments. We interacted the affordability of the housing proposal with the amount of compensation offered, and show these results in Figure 6. For affordable proposals, the compensation offered had no influence on respondents’ support (the coefficient represented by the filled circle at the bottom of Figure 6). Yet for market-rate housing proposals (the coefficient represented by the filled square in Figure 6), the amount of compensation offered increased respondents’ support.Footnote ¹⁰ In substantive terms, a 100 percent increase in compensation increased support for market-rate proposals by 3.7 percentage points (p < .001), but the same increase in compensation did not increase support for affordable proposals.Footnote ¹¹

Figure 6.

Predictors of support for housing proposals, and affordability interacted with compensation. Lines indicate 95%-confidence intervals (thin lines) and 90%-confidence intervals (thick lines).

Evidence from open-ended responses

To better understand why the effect of compensation varied based on the affordability of the proposed housing development, we turned to the open-ended text responses given by survey respondents. For the first of the five proposals viewed, we asked each respondent: “Using at least 5 words, how did the financial compensation affect your support for the proposal?” We calculated the frequency of words that people used in response to this question, among both those who were randomly assigned a proposal that included affordable housing and those assigned a proposal solely composed of market-rate housing. To make responses comparable, we stemmed all words, removed numbers and stopwords (i.e. common conjunctions and prepositions), and replaced the symbol “$” with the word “dollars.”

To compare the responses of people who evaluated affordable proposals with those of people who evaluated market-rate proposals, we examine the relative frequency of each word following the approach used by Wasow (Reference Wasow2020) and others. Figure 7 presents the relative frequency of words used by respondents in the two conditions, among the most commonly used (overall) words.Footnote ¹² We plot the base-2 logged ratio of a given term’s frequency between respondents in the affordable condition and respondents in the market-rate condition along the horizontal axis. Positive values of this ratio indicate words that are more common among respondents in the affordable condition, while negative values of this ratio indicate words that were more common among respondents in the market-rate condition. Of course, using word stems alone cannot indicate whether a reference to a word was positive or negative. We can only conclude that the stem was more frequently used and therefore likely more salient in respondents’ decision-making process.

Figure 7.

Log2 ratio of term frequencies in open-ended text comments regarding financial compensation (affordable housing treatment/market-rate housing treatment).

The positive values of the relative frequency of the top two words in Figure 7 indicate that respondents evaluating affordable proposals were more than one-and-a-half times more likely to use the terms “benefit” and “build” compared to those respondents evaluating market-rate proposals.Footnote ¹³ In contrast, the bottom two words in Figure 7 indicate that respondents evaluating market-rate housing used the term “afford” almost three times as much as those evaluating an affordable housing proposal, and referenced the compensation offered to them (using the word “dollar”) almost twice as frequently.Footnote ¹⁴

Respondents who evaluated market-rate proposals appear much more focused on the compensation offered by developers than respondents who evaluated affordable proposals. Respondents evaluating affordable housing may have paid more attention to other features of the proposed development – such as whether the housing itself would benefit the community – instead of the compensation. This evidence provides depth to our earlier results showing that the amount of compensation only minimally influenced respondents’ support when they were evaluating affordable housing proposals.

Policy implications

The goal of our experiment was to assess how several features of housing proposals influence local residents’ support for those developments. While our experiment uses real locations, real images of buildings, and renderings from respondents’ self-reported neighborhoods, our study is unable to fully replicate the debate that accompanies new housing proposals: experiments inherently require simplification and abstraction. Despite this abstraction, we believe our experiment has value in that it allows us to learn about people’s reactions to developments at their first chance to give input and decide whether they support or oppose development – a crucial beginning from which people may eventually enter the more public discourse. In this section, we discuss the limits to the implications of our findings for policymaking given the constraints of external validity due to experimental simplification.

First, the real-world approval of housing development rarely takes the form of binary votes on a neighborhood-wide ballot. Instead, negotiations are often held between developers and select community groups. To the average voter, so long as “the community” gets something, the proposed package may be seen as adequate. In contrast, politically powerful local groups are likely to care deeply and attempt to shape the type of community benefit offered. If the public benefits in our experiment were more selective (semi-public, semi-excludable), then it is possible that respondents would show more support for the personal direct payments. Likewise, the public may be skeptical of community benefits if the negotiation process were framed as illegitimate. Ultimately, these questions and scenarios underscore how little research has been conducted on public support for compensation in the context of housing.

Second, while we report average treatment effects, there is variation in these effects across relevant real-world groups of people. Table H-11 and Table H-12 in the Appendix show the findings from our main analyses, separated by survey waves. Wave 2 is composed of respondents recruited from neighborhood organizations who are active in housing politics, and may therefore bear the most relevance for the real-world development process. Respondents in this sample appear more supportive of housing which is farther away from them. Likewise, Wave 2 respondents also show a smaller increase in support for housing that includes affordable units compared to solely market-rate housing than among our full respondent sample. The effect of compensation is also smaller for respondents from Wave 2 and not statistically significant.

These differences across sample waves affirm expectations of differences across residents. Those who are the most heavily invested in neighborhood politics appear to fit the typical definitions of housing NIMBYism: opposed to proposals most nearby and wary of affordable housing. At the same time, the treatment effects are still relatively consistent across samples, suggesting that our study is detecting real attitudes held among residents and which may be channeled using policy innovations like we propose. These subsample analyses were not preregistered, so we caution against over-interpretation of these results. Yet they give some clues as to how dimensions of housing development that we examine may play out in an especially important subset of the population, and may be useful and worthy of future study.

Third, as we state in the Discussion, the positive effect of affordable housing on respondent support may be driven by the type of affordable developments proposed: mixed-income. But affordable housing can take many forms, even in the single city of Boston. New larger developments covered by Boston’s inclusionary zoning policy only need to dedicate 13% of their units to income-restricted housing. But in Massachusetts since 2010, new construction funded by the Low-Income Housing Tax Credit has designated 96% of its units as income-restricted. Our proposals, which set aside 50% of units as income-restricted, fall somewhere in between these two extremes. We believe these proposals tap into the attitudes towards mixed-income housing and should not necessarily be interpreted as relevant to either fully affordable or inclusionary zoning developments.

Fourth, compensation agreements may operate differently outside the context of Boston. But as we show in our observational analysis, community contact and informal negotiation is a mandatory part of housing entitlement process in 12 of the 25 most populous cities in the U.S. Likewise, many local policy advocates point towards compensation via community benefits agreements as the more equitable path to development (Fraser Reference Fraser2022) – not just of housing, but also development of energy infrastructure (U.S. Department of Energy 2017). While CBAs may be not be ubiquitous in housing today, our findings are a necessary first step to better understand the political science of compensation as well as the policy conditions under which compensation will be effective in winning public support.

Even with these caveats, our findings indicate clear pathways toward increasing public support for new housing in urban environments such as Boston. Our results suggest that including affordable housing can be a useful measure to increase net support for a project. However, once affordable housing is incorporated, additional compensation is unlikely to prove useful in expanding a coalition. In fact, additional compensation may only hurt the financial viability of a project, with little payoff in public support. Instead, support for mixed-income developments, which are often developments that fall under commonly used inclusionary zoning requirements, could be increased by highlighting the relative affordability of the development. This appeal to the symbolic value of affordability may do more to garner neighbors’ support compared to simply increasing financial compensation.