Impact of large-scale solar on property values in the United States: Diverse effects and causal mechanisms

Significance

Large-scale solar projects are crucial for decarbonizing the US economy, but growing local resistance may impede the renewable energy transition. We estimate the impact of large-scale solar on property prices and the underlying pathways using 8.8 million sales and 3,699 solar sites in the United States. Exposure to solar sites decreases nearby residential home values but increases land values. For large-lot homes, the increase in land value largely mitigates the negative residential impact. Varying county political leaning and land use histories result in significantly different residential value impacts. Empirical evidence indicates that the current negative residential impact might represent a stigma effect attached to solar sites. Our findings provide important insights for addressing local resistance against large-scale solar projects.

Abstract

As the renewable energy transition continues into less receptive communities, local opposition is expected to intensify, potentially slowing the process. Since the local impacts are neither well quantified nor widely recognized, we lack policies and common practices to mitigate the potential associated welfare loss in affected communities. Based on a nationwide dataset combining property transactions and large-scale solar photovoltaic (LSSPV) sites, we analyze the heterogeneous effects of LSSPV on property prices and the associated causal pathways. Difference-in-differences estimates show that LSSPV significantly increases agricultural or vacant land value by about 19.4% within a 2-mile radius, while simultaneously reducing residential property values within 3 miles by about 4.8%. The estimated average negative impact on home values is primarily driven by site proximity and diminishes with both distance and time. Effect estimates are more robust to alternative specifications when proximity pairs with visibility rather than invisibility, but no evidence suggests visibility significantly amplifies the proximity effect. Heterogeneous effect estimates indicate that high solar lease potential, being in heavily Democratic-leaning counties, and brownfield redevelopment largely mitigate the negative residential value impact. The analysis reveals no significant heterogeneity across a few factors, including varying site visibility, directional orientation of properties relative to the LSSPV site, and different tracking systems. Evidence indicates that the negative impact on residential values might mainly stem from negative perceptions, but channels through physical conditions cannot be entirely dismissed. Our assessment provides benchmark information for local externality mitigation plans, potentially reducing community opposition and expediting the renewable energy transition.

As the cost of solar energy continues to decline (1), solar is likely to remain the leading source of renewable energy in the United States (2). Although the climate benefits of large-scale solar photovoltaic (LSSPV) are widely recognized, the siting of LSSPV projects has encountered increasing local resistance (3–5). As the renewable energy transition deepens into less receptive communities, local opposition is expected to intensify and slow down the transition process. Anecdotal and qualitative evidence suggests that the local concerns are primarily driven by negative aesthetic impacts, decreased property values, environmental injustice, and adverse impacts on local agriculture (3, 6, 7). However, these negative impacts are not well quantified, and we lack policies or common practices to mitigate the potential welfare loss in affected communities.

LSSPV facilities can significantly alter local amenities in residential areas. Recent studies suggest that proximity to a solar site may reduce home values (8, 9) due to diminished amenities such as adverse visual impact (10). The geometric and highly reflective surfaces of LSSPV facilities can be seen as unattractive and disruptive, particularly in natural or agrarian settings (11). There are other potential disamenities associated with LSSPV that may not be revealed immediately after site installation, including disrupted ecosystems and wildlife habitats (12, 13), increased soil erosion and water runoff, and degraded air quality (14). Moreover, negative perceptions of disamenities could lead to property value losses that are unrelated to actual levels of physical disamenities, a phenomenon known as the stigma effect in the housing market (15, 16).

Solar development can affect land prices considerably. An LSSPV facility typically requires between 5 and 10 acres per MWac of generating capacity. Agricultural land has been the most common land type for LSSPV development, due to its suitability, such as being flat, dry, cleared of natural vegetation, and close to electric infrastructure (17, 18). A recent projection from the American Farmland Trust shows that solar projects could occupy over 7 million acres by 2040, with 83% of new installations on farmlands and ranchlands, half of which are on highly productive land (19). If we consider the potential future surge in energy demand, e.g., electrifying the transportation sector and establishing AI data centers, the required farmland for solar energy production could be much higher than the projected 5.8 million acres. In the long run, the land use competition between solar development and agricultural production is likely to increase the scarcity of farmlands, especially at the urban fringe. In the short run, leasing the land for solar energy production provides higher financial returns than traditional agricultural operations, which may drive up farmland prices and elevate farming costs.

Existing studies provide suggestive evidence that visual impacts and loss of property values are the two leading concerns for local oppositions (3, 10).^* These local impacts of LSSPV represent classically defined externalities, as no widely established mechanism exists for solar site owners to compensate neighboring communities for potential negative effects. Quantifying these externalities is important to establishing solar siting procedures that adequately compensate the community and allow socially optimal allocations of resources. More importantly, as solar sites are initially developed in receptive communities, siting efforts are expected to become more challenging when the renewable energy transition continues. Studies have suggested that a major proportion of proposed LSSPV projects were denied or withdrawn due to local resistance (5, 10).^† Clarifying and addressing the externalities of LSSPV development will help alleviate local opposition to solar development and accelerate the energy transition.

Utilizing property-level transaction data and detailed LSSPV site information, we present a comprehensive nationwide analysis to estimate and quantify the externalities of LSSPV facilities facing nonresidential and residential properties. We employ a Difference-in-Differences (DID) identification framework to investigate the effects of solar projects on nearby property values. Previous studies have employed similar methods to investigate the property value effects of solar site exposure in a few selected states (8, 9, 20). While viewshed analyses and visual impact investigations are prevalent for wind site studies (e.g., refs. 21–25), previous solar studies have not measured site visibility or quantified the associated visual impact, despite some indicating its relevance (e.g., ref. 26). In contrast to previous solar studies focusing on site proximity, we additionally assess the impact of site visibility and its interaction with proximity. Specifically, we create a geospatial database showing the visibility from every residential home to nearby LSSPV facility in the contiguous United States (Fig. 1, see Data and Methods for details). With the average effects showing the general size of welfare changes in the neighborhood, we further differentiate the impact mechanisms and provide information for a compensation plan for the local externalities generated by LSSPV sites.

Fig. 1.

Map of LSSPV locations, capacity, and visibility. The size of circles indicates the capacity of each LSSPV site. The colors represent the visibility of each site. Visibility is measured in the number of local (<6 miles) residential homes with a view of that LSSPV site.

Our analysis demonstrates that LSSPV sites affect local residential property values and land values differently. We separately analyze transactions on three types of properties. The first type is residential properties (hereafter “residential homes” or “residential”) with a lot size under five acres (i.e., the typical minimum acreage requirement for a solar lease), where LSSPV effects primarily stem from impacts related to residential amenities. The second type involves agricultural or vacant land above five acres (hereafter “agricultural land” or “ag-land”), where LSSPV effects mainly result from potential solar lease-induced land use value changes. The third type includes properties over five acres with residential structures (hereafter “large-lot homes”), where LSSPV effects may include both residential amenity and land use value impacts. Within the analysis of each property type, we further investigate the impact heterogeneity across a range of dimensions, including rural–urban status, census region, lot size, county political leaning, median household income, solar site scale, site historical land use, state siting regulation, among others. To make sure our estimates are not specific to the five-acre segregation criterion, we conducted robustness checks in SI Appendix.^‡

1. Results

1.1. LSSPV Impact on Residential Home Value.

We first present the results for residential properties under five acres, which include approximately 8.3 million property transactions within a 6-mile radius of LSSPV sites from 15 y before the installation of each site through 2020. Further analytical details are provided in Data and Methods.

1.1.1. Residential proximity and visibility.

We first use distance decay specifications within the DID framework (see Section 3.5 for model details) to decide the proper treatment variable, assuming solar site exposure is determined by proximity and visibility. The view-specific distance decay results (Fig. 2) show that proximity is the major driver of the negative residential value impact. We find that, without LSSPV view, LSSPV proximity reduces residential sales price by up to 7.2% within a 0.5-mile radius, and the bin-specific estimates gradually decrease with distance and remain statistically significant up to 3 miles from the LSSPV site. Having LSSPV in the viewshed of a home incurs slightly more negative effects (i.e., up to 7.9% within 0.5 miles) compared to the pure proximity effects,^§ and the bin-specific effects also diminish with distance. Beyond 3 miles, both the proximity effect and the visibility effect become indistinguishable from zero, suggesting that visibility does not independently generate negative impacts in the absence of proximity.

Fig. 2.

Effects of proximity and view on residential home value. The blue line connects the coefficient estimates of proximity bins without view, obtained by interacting the proximity bins, the binary posttreatment indicator, and the no-visibility indicator (i.e., equals 1 if no site view). The red line connects coefficient estimates of proximity bins with view, obtained by interacting the proximity bins, the binary posttreatment indicator, and the visibility indicator (i.e., equals 1 if with site view). The number of observations (N) in this analysis is 8,303,074, excluding singleton observations on the census-tract by year level. The 95% CIs are constructed with two-way clustered SEs at the census tract and year level. The control group is properties in the 5-to-6-mile proximity bin.

1.1.2. Residential treatment—site within 3 miles.

As shown in Table 1 column (1), when examining the average treatment effect of proximity within 3 miles (regardless of visibility), the estimate is 4.8% and statistically significant at the 5% level. We further investigate the interaction between proximity and visibility in column (2). When the solar site is visible and within 3 miles, property values, on average, decrease by about 5.2%. The corresponding effect of an invisible site is estimated at 4.6%. While both estimates are statistically significant at the 5% level, a statistical test shows that the difference between them is not significant at all (test P-value = 0.746), indicating that site visibility may not impose a significant additional average effect beyond proximity and supporting the validity of proximity-based specifications in prior studies (e.g., refs. 8 and 9). We also checked an alternative specification that excludes no-view properties within the 3-mile radius in column (3) of Table 1, which provides a similar interaction effect of visibility and proximity. These average effect analyses, combined with the distance decay results, suggest that site proximity alone largely drives the residential home effect. Consequently, site proximity within a 3-mile radius [as presented in Table 1 column (1)] serves as the principal treatment variable, representing LSSPV exposure, in subsequent event study and heterogeneity analyses. Examining the sensitivity of estimates to alternative control group specifications in SI Appendix, Table S6, we find that the interaction effect of visibility and proximity remains robust across the board, while the pure proximity effect becomes insignificant in some of the alternative specifications. This implies that site visibility appears to reinforce the proximity effect in the sense that it improves the robustness of the home value effect estimate across various alternative control group specifications. To provide a comprehensive view of the proximity effect, we present both the specifications from column (1) and column (2) in the pretrend tests and robustness checks in the SI Appendix. Pretrend tests with placebo treatments in SI Appendix, Table S5 show that the parallel trend assumptions are satisfied for all specifications in Table 1. More robustness checks in SI Appendix, Tables S8 and S9 confirm that all estimates in Table 1 remain consistent when applying alternative sample selection criteria based on acreage and the number of observations per tract-year cluster.

Table 1.

DID Estimates for Residential Homes

	(1)	(2)	(3)
	ProxT	ProxT × ViewT	ProxT × ViewT
ProxT	−0.076**
	(0.022)
${\beta }_3$: ProxT × Post	−0.048*
	(0.020)
ProxT × 0.ViewT		−0.078**
		(0.022)
ProxT × 1.ViewT		−0.070**	−0.044
		(0.023)	(0.029)
${\beta }_3^{no\_view}$: ProxT × 0.ViewT × Post		−0.046*(0.020)
${\beta }_3^{\mathit{view}}$: ProxT × 1.ViewT × Post		−0.052*	−0.046+
		(0.020)	(0.023)
N	4975808	4975808	2444983
Covariates	Yes	Yes	Yes
Census Tract × Year	Yes	Yes	Yes
Test ($H_0$: ${\beta }_3^{no\_view}={\beta }_3^{\mathit{view}}$): z-Statistic = 0.324 P-value = 0.746

Note: In Column (1), ProxT, standing for site proximity below 3 miles, is used as the treatment. In Column (2), proximity without view (ProxT×0.ViewT) and proximity with view (ProxT×1.ViewT) are used as treatment. In Column (3), properties that satisfy ProxT = 1 and ViewT = 0 are excluded. ${\beta }_3$ s represent the treatment effects specified in Section 3.5.1. SE, two-way clustered at census tract and year level, are reported in parentheses: ⁺P < 0.1, ^∗P < 0.05, ^∗∗P < 0.01, ^∗∗∗P < 0.001. Census tract by year fixed effects and property-level covariates are included in all specifications but not displayed. The control group is properties in the 5-to-6-mile proximity bins of the LSSPV sites, and properties located within 3 to 5 miles from the LSSPV sites are excluded. The number of observations, N, is calculated excluding singleton observations on the census-tract by year level. The coefficient for Post is omitted due to collinearity with fixed effects.

1.1.3. Residential event-study results.

We explore the timing of the LSSPV exposure effect (i.e., site visible within 3 miles) based on an event study where the base year is specified as 3 y prior to the LSSPV installation^¶ (Fig. 3). The average negative price impact on residential homes is minor after the base year but becomes pronounced following the installation. The effect generally maintains its magnitude over time and fades after the ninth year postinstallation. There are potential explanations for the observed effect dynamics. Right after the base year, the gradual dissemination of the LSSPV site information may not have reached many home buyers or led them to fully realize the potential negative price impact of the site, but the installation event makes the impacts clear and manifested in the market. The diminishing effect after 9 y might come from the shrinking sample size as most of the LSSPV sites were developed after 2010. However, if the diminished effect is true, it does not necessarily imply that the negative amenity impacts disappear after 9 y since many of the negative impacts, such as soil erosion and dust pollution, may take a long time to manifest (14, 27, 28). A more plausible explanation of the faded price impact may be linked to residential sorting and demographic shifts (29–31), as individuals less concerned about LSSPV facilities move into the affected neighborhoods. This indirectly suggests that the negative price impact might be more closely related to psychological factors than to the amenities themselves, which will be explored further in subsequent analyses and discussions.

Fig. 3.

Event study on residential home value. The treatment (LSSPV site within 3 miles) effect on residential home values is illustrated across different years relative to the year of LSSPV installation. The blue squares on the black line indicate the coefficient estimates, obtained by interacting the treatment variable with year indicators. The reference year is defined as 3 y before the LSSPV installation, and the control group is properties in the 5-to-6-mile proximity bin. The shaded areas represent the 95% CIs, constructed using two-way clustered SEs at the census tract and year level.

1.1.4. Residential Effect Heterogeneity.

We explore the heterogeneity of LSSPV exposure effect on residential homes across various dimensions, as shown in SI Appendix and Fig. 4.^# We observe noticeable heterogeneity across census regions, county political leaning, county median household income, and historical land use of the LSSPV sites. Statistical tests results are available in SI Appendix, Table S10. LSSPV sites in the Northeast region impose significantly more negative impacts than those in other regions. Heavily Democratic-leaning counties (over 65% Democratic votes in 2016) experience a positive LSSPV effect (+0.0374, insignificant), which is significantly different from more politically conservative counties (−0.0538, significant at the 5% level). Greenfield LSSPV development leads to a negative effect (−0.0466, significant at the 10% level), while brownfield redevelopments lead to a positive residential value effect (+0.225, significant at the 10% level), significantly different from the effect of Greenfield LSSPV.^|| Observed differences along other dimensions are not statistically significant. Moreover, we observe almost zero heterogeneity across different rural status, different lot sizes, different site capacities, and different levels of site visibility. A higher level of visual exposure (“High View” in Fig. 4) or directly facing the solar panels (i.e., in the south of the solar panels, “Facing” in Fig. 4) does not lead to a more negative residential value effect, providing further evidence that more view exposure may not lead to significantly more negative impacts. While we lack direct data on glint and glare effects, indirect evidence suggests they may not be a primary mechanism, as we find no evidence to support that being exposed to a site with tracking systems (i.e., potentially more susceptible to glare impacts, “Tracking” in Fig. 4) or facing the solar panels lead to more negative impacts. Instead of visual levels or details, impacts appear to stem from psychological factors, such as negative perceptions of industrialization and altered scenic views. These negative perceptions are expected to be amplified by conservative ideology or mitigated by progressive ideology, aligning with the empirical finding that more politically conservative counties are associated with more negative impacts.

Fig. 4.

Heterogeneous effects of LSSPV exposure by different dimensions. Diamonds are the point estimate of the effect of LSSPV on nearby residential home values based on DID models. The treatment is LSSPV within 3 miles, and the control group is properties in the 5-to-6-mile proximity bin of the LSSPV site. The 95% CIs of the estimates are shown as bars, having clustered SEs at the census tract and year level. Check SI Appendix for the details of all factors investigated here. More heterogeneity checks differentiating visible and invisible sites are available in SI Appendix, Fig. S7.

1.2. LSSPV Impact on Agricultural Land Value.

Our Ag-land analyses show that having LSSPV sites within 2 miles of agricultural or vacant land increases the sales price per acre by an average of 19.4%,^** which is statistically significant at the 5% level (Fig. 5). The positive effect rapidly declines and becomes insignificant beyond 2 miles, similar to estimates in ref. 32. This positive effect is likely due to the demand increase from potential solar leases, as further expansion of existing LSSPV sites is less costly than constructing new sites and likely involve nearby agricultural or vacant land. Pretrend tests in SI Appendix, Table S5 show that the parallel trend assumptions are satisfied. Robustness checks in SI Appendix, Table S7 suggest that our main ag-land estimate is robust against different control group selection criteria. Event-study results in SI Appendix, Fig. S5 show that the positive land value effect manifests 3 y after the site installation and fades away 6 y later. SI Appendix, Fig. S6 presents our analysis of heterogeneous ag-land effects. We find that LSSPV sites of larger than-median scale have virtually zero effect on land value, while sites of smaller scale display a positive effect on land value (significant at the 10% level). Considering that smaller sites have a larger potential for expansion, this observation seems to confirm our speculation that the nearby land value increase is mainly driven by the potential of future solar lease. We also find that agricultural or vacant lots of large acreage bear virtually zero effect while smaller lots show a significantly (at the 5% level) positive effect. However, these differences are not statistically significant. More robustness checks in SI Appendix, Table S8 suggest that our land value estimates remain consistent when applying alternative sample selection criteria based on acreage. Finally, robustness checks in SI Appendix, Table S9 reveal that when focusing solely on county-site-year clusters containing more than a few sales, the land price effect of LSSPV rises dramatically, reaching 86.1% when excluding less-than-20-land-sales clusters (corresponding to a coefficient of 0.621). Given that we have excluded sales of land hosting LSSPV sites, the mechanism behind this substantial effect on land prices remains unclear but warrants further investigation.

Fig. 5.

Distance decay results for agricultural/vacant land and large-lot homes. The top subfigure shows estimates for agricultural and vacant land above five acres. The bottom subfigure shows estimates for large-lot homes, defined as properties over five acres with residential structures. The results show the value effects of LSSPV for a range of proximity bins, defined with 2-mile intervals. The blue line connects the coefficient estimates of proximity bins, obtained by interacting the proximity-bin indicators with the binary posttreatment indicator. The treatment groups are properties within these proximity bins, while the control group is properties within the 18-to-20-mile proximity bin. The 95% CIs are constructed with two-way clustered SEs at the county-site and year level.

1.3. LSSPV Impact on Large-Lot Home Value.

Our empirical results show that LSSPV sites have a dual effect: they decrease residential property values via reduced residential amenity, while simultaneously increasing nearby land prices due to enhanced land use potential. For large-lot residential homes with over five acres of land, we expect the LSSPV to impact property values through both channels. Our distance decay analysis (Fig. 5, Bottom) suggests that the overall LSSPV impact on large-lot home price is close to zero and statistically insignificant for all nearby proximity bins. Robustness checks in SI Appendix, Tables S8 and S9 confirm that these large-lot-home estimates remain small and insignificant when applying alternative sample selection criteria based on acreage and the number of observations per tract-year cluster. Therefore, the LSSPV property value impacts via amenity reduction and increased land use potential seem to offset each other in residential homes with over five acres of land.

2. Discussion

This study provides a comprehensive nationwide assessment of the externalities associated with LSSPV installations in the United States focusing on their impacts on property values. We leverage a rich property transaction dataset with detailed geospatial information of LSSPV sites to estimate the effects on both residential properties and agricultural/vacant land. We apply advanced geospatial methods to overcome computational challenges and develop a comprehensive nationwide database on LSSPV visibility. Our findings reveal that LSSPV installations negatively affect the value of residential properties located within 3 miles, while increasing prices for agricultural and vacant land within 2 miles. Moreover, when the impacts through reduced residential amenity and increased land use potential coexist, the LSSPV effect on large-lot homes is indistinguishable from zero. We also explore the dynamics and heterogeneities of the local property value effects of LSSPV.

Our analyses and heterogeneity checks indicate that a nearby solar site may act as a stigmatizing nuisance (i.e. a psychological disamenity, see refs. 15, 16, 33) and 34). Evidence supporting this claim includes the minimal variation in effects across different levels of site visibility, in effects across properties to the south and to the north of the site, and in effects across sites with different tracking systems, as they suggest that the view details of solar sites (including view extent, the exact view composition, and potential difference in glare effects) do not significantly impact residential values. The negative impact on nearby residences appears to operate primarily through psychological channels rather than through the degree of visibility or specific visual details. Considering disamenities other than visual impact, the scale of the site likely results in different disamenity levels and impacts, but this is also not observed (i.e., “Big USS” vs. “Small USS” in Fig. 4). One explanation can be linked to negative perceptions that solar sites are industrial/commercial uses that alter rural land use and scenic views (15). The disparities in effects between brownfield and greenfield sites align with this mechanism. Another piece of evidence is the significantly higher property value loss in more conservative counties compared to Democratic-leaning counties. This disparity is likely due to solar sites being more aligned with progressive values prevalent in Democratic-leaning counties and less frequently associated with negative perceptions. However, we cannot entirely rule out causal channels related to actual disamenity variations. First, our nationwide analysis may obscure heterogeneities under certain conditions – for example, sites with a larger scale may have a stronger negative effect in the Northeast but a weaker one in the West, potentially canceling out in a pooled sample. Second, unexplored physical channels, such as vegetation and soil management practices (e.g., refs. 9 and 27), might also contribute to the negative LSSPV impact on residential values.

Our findings highlight the complex interplay between the benefits and costs of LSSPV development. In SI Appendix, Table S11, we performed a back-of-the-envelope calculation to estimate the benefits and costs of LSSPV solar sites included in our analysis, including the mitigation value (i.e., avoided social cost of carbon emission), the appreciation of nearby agricultural or vacant land value, the value loss of nearby residential properties, and the agricultural production loss on land utilized for hosting LSSPV. The results suggest that the assessed benefits of existing LSSPV significantly outweigh the assessed total costs. The carbon mitigation benefit is the major benefit (about $22.2 billion annually), while the loss in residential home value is the dominant cost (about -$4.1 billion annually). Therefore, property value losses constitute a major proportion of negative externalities of LSSPV. While the expansion of solar energy is crucial for the renewable energy transition, it is imperative to address the localized externalities to ensure equitable outcomes for affected communities. Quantitative evidence, such as that generated by this study, can inform policymakers and stakeholders in designing compensation mechanisms and siting strategies that mitigate negative impacts while promoting the broader adoption of solar energy.

To illustrate how our results or similar studies could be used to develop a community compensation plan, we design a prototype evidence-based community compensation plan for a site proposal in (SI Appendix, Fig. S12). First, property value impact studies should be carefully conducted with empirical data from comparable solar sites (e.g., similar size, similar demographics, in counties or states of similar regulations, etc.), where the effect of distance decay, dynamics, and heterogeneities across a wide range of dimensions should be analyzed. The sample choice of LSSPV sites needs to balance site similarity and statistical power of analysis. Second, based on the property value study, compensation specifics should be decided for different properties in the neighborhood. Taking our main results as an example, compensation rates could be set at 5.2% of the annualized property value for 10 y for residential homes within 3 miles of the LSSPV site with a site view, 4.8% for those without a view, and 19.4% of annual agricultural land rental costs^†† for 4 y for leasing farmers within 2 miles. Third, the community compensation plan can involve communication with stakeholders ahead of the permitting process, and stakeholders’ input should be involved in the revision process before reaching a final plan. A comprehensive compensation plan should also consider local externalities that might not visibly manifest in property prices. We would like to stress that the specific community compensation plan developed based on our nationwide study here should be merely taken as an example, and we recommend conducting targeted studies to determine appropriate community compensation plans for a specific LSSPV site.

3. Data and Methods

The analysis primarily utilizes data of three categories: The US LSSPV data, the real estate transaction and assessment records, and geospatial data.

3.1. LSSPV Data.

The LSSPV data acquired from the US Large Scale Solar Photovoltaic Database (USPVDB) (35) contain 3,699 LSSPV facilities investigated in the study. This dataset provides detailed information on LSSPV site footprint, area, capacity, and installation year, spanning from 1986 to 2021 (SI Appendix, Fig. S1 shows the total acreage developed per year, and SI Appendix, Table S1 shows the summary statistics of LSSPV projects). The facility polygons are digitized along the boundaries of the solar arrays, within an accuracy of 10 m.

3.2. Property Transaction.

The property data are purchased from CoreLogic through a data agreement. CoreLogic data contain comprehensive information on property and transactions from the whole United States and enables researchers to work on property-level research questions. We developed a process to exclude non-arm’s-length transactions (i.e., purging price outliers, foreclosure sales, multiple sales, sales between relatives, sales involving institutional buyers or sellers, and others as detailed in SI Appendix) so that our analyses only include transactions reflecting fair market values. The transaction prices are adjusted for inflation to reflect their values in 2017 dollars using the Consumer Price Index data from the US Bureau of Labor Statistics. We also exclude potential home flipping events by removing transactions of the same property that occur within 120 d of each other. As the majority of LSSPV sites have been developed within the past decade, we keep transactions up to 15 y before the installation of nearest LSSPV to make the time frame generally centered around the LSSPV development. The final dataset for analysis comprises both single-family residential properties and agricultural or vacant land, spanning 40 states^‡‡ from 1993 to 2020. To avoid the potential impact from market disequilibrium, we drop observations during the Great Recession (i.e., 2008 to 2010). SI Appendix, Tables S2–S4 show the summary statistics of residential homes, agricultural and vacant land, and large-lot homes, respectively. SI Appendix, Figs. S2–S4 illustrate the distribution of post-LSSPV-installation transactions of residential homes, agricultural or vacant land, and large-lot homes, respectively, across different proximity bins.

3.3. Geospatial Data.

The geospatial data consist of a collection of geographic layers obtained from the US Census Bureau TIGER/line geodatabase (USCB TIGER) and US Energy Information Administration (EIA), which includes shapefiles of primary roads, transmission lines, and metropolitan areas. To support heterogeneity analyses, we also collected data on median household income, median land values, political leanings, and state-level siting policies, among other factors (see SI Appendix for details).

To acquire solar site proximity and other (dis)amenities, we generated geographic variables that represent the Euclidian distance between a property and the boundary of the nearest five solar sites, transmission line, primary road, and metropolitan area. The geographic variables were then matched with the property data. To alleviate identification concerns that attributes of control observations (i.e., properties far away from sites) might considerably deviate from treated observations (i.e., properties with solar site exposure), we only kept residential homes that are less than or equal to 6 miles away from the nearest solar sites. For properties above five acres (i.e., agricultural land or large-lot homes), we use a 20-mile radius inclusion criterion due to the general low density and low transaction volumes of such properties. The final sample includes 8.3 million transactions for residential homes, 68 thousand transactions for agricultural or vacant land, and 416 thousand transactions for large-lot homes.

3.4. Visibility Analysis.

We establish a visibility database for LSSPV across the continental United States and investigate the property value effect of LSSPV visibility. We calculate the visibility from residential properties to large-scale solar sites within 6 miles. This visibility analysis proceeds in three steps. First, we acquire Digital elevation models (DEMs) of the continental United States from the Shuttle Radar Topographic Mission (SRTM) produced by NASA.^§§ Our analysis uses the 2018 version of SRTM DEMs at a resolution of 90 m by 90 m. The DEMs employed reflect terrain elevation but may not capture structures (e.g., houses or trees), and hence could overstate visibility especially when the viewpoint and the target are close (36). Nonetheless, the employed DEMs are the best available public data for our analysis, as structural elevation data (e.g., Light Detection and Ranging, or LiDAR, data) are not available for most solar sites and their neighborhoods.

Second, we calculate the viewsheds from solar sites to decide the areas from which the sites are visible, utilizing the duality of vision following ref. 21 (i.e., if and only if viewpoint A has a view on target B, a viewpoint on B has a view on target A). This approach greatly reduces computational effort since the number of solar sites (3,699) is much smaller than the number of properties (about 5.9 million). Unlike the wind turbines that require height specifications for accurate viewshed analyses, LSSPV sites span broad areas, necessitating a proper way to account for partial views of a large solar site. Specifically, we set viewpoints along the perimeter of each site, where the viewpoints are defined with a random start point, an interval distance D, and a height of two meters. In practice, D is set at 500 m to balance the computation workload and the accuracy of partial view accounting.

Third, we aggregate the viewsheds from all site perimeter viewpoints and overlay the aggregated viewshed layer with properties to calculate the visibility variables. The aggregation of viewsheds will generate the visibility index (Fig. 6) for each geographic unit defined by the raster resolution (90 m by 90 m). Overlaying with the property layer, the visibility index will represent the number of perimeter viewpoints that can see a property, or the number of solar site perimeter points that the property has view on based on the duality of vision. This property-specific visibility index quantifies the extent of solar site visibility for each property and can be converted into a binary visibility variable that serves as the treatment variable in a DID model. For more details of the visibility analysis, refer to Visibility Analysis Details section in SI Appendix.

Fig. 6.

Surface of Visibility Index. The visibility index measures the number of visible perimeter points of nearby solar sites. Intuitively, the red color denotes regions with solar view, and regions in darker red can see a larger area of solar panels.

3.5. Econometrics: Property Value Effect Models.

Previous studies have used econometric models to analyze and identify a variety of characteristics that could consistently influence property values, such as the productivity of the farmland (e.g., ref. 37), the influences of urbanization (e.g., ref. 38), and environmental factors (e.g., refs. 39–41). To estimate the impact of solar projects on nearby property values, it is crucial to control for potential confounders. We employ a DID approach to investigate the effects of LSSPV installation on nearby property values. Intuitively, this approach compares the change in property values before and after installation for properties close to the LSSPV site against the value change for properties farther away but still within the defined vicinity.

3.5.1. Analyses for residential homes.

The general DID framework of our residential home study is as follows:

$$\begin{array}{c}\text{ln}\left(P_{\mathit{it}}\right)={\beta }_0+{\beta }_1{\mathit{Post}}_{\mathit{it}}+{\beta }_2{T}_i+{\beta }_3{\mathit{Post}}_{\mathit{it}}\times T_i\hfill \\ +{\delta }_k{\displaystyle \sum _{k=1}^K}{X}_{\mathit{it}}^k+{\gamma }_k{\displaystyle \sum _{k=1}^K}({\mathit{Post}}_{\mathit{it}}\times X_{\mathit{it}}^k)+{\tau }_{\mathit{ct}}+{\epsilon }_{\mathit{it}}.\hfill \end{array}$$ [1]

In Eq. 1, each observation corresponds to a transaction of residential home $i$ that occurred in year $t$, with the dependent variable being the natural logarithm of transaction price $\ln \left(P_{\mathit{it}}\right)$. ${\mathit{Post}}_{\mathit{it}}$ is a binary indicator that denotes whether the transaction of residential home happened after the LSSPV installation. $T_i$ is the binary indicator that denotes whether a residential home was assigned to a treatment group, and the exact definition of treatment is explained below. The coefficient ${\beta }_3$ associated with the interaction term between ${\mathit{Post}}_{\mathit{it}}$ and $T_i$ captures the impact of LSSPV installation on the outcome variable, which resembles a proportional change in the residential home prices. Previous studies show that the proximity to transmission lines could have an impact on the value of nearby property (42), and this impact could change after an LSSPV installation in the vicinity (20). To account for housing and lot characteristics that could affect home values and the estimation of ${\beta }_3$, we include property-level control variables $X_{\mathit{it}}^k$ and ${\mathit{Post}}_{\mathit{it}}\times X_{\mathit{it}}^k$ (43, 44), where $X_{\mathit{it}}^k$ include total bedroom number, total bathroom number, building age, and natural logarithms of distances to the nearest transmission line, the nearest primary road, and the nearest metropolitan area. To absorb the time-varying external location-specific shocks in the housing market, we incorporate fixed effects on the census tract by year level, denoted as ${\tau }_{\mathit{ct}}$. All SE are two-way clustered at the census tract and year level.

To detect the proper site-proximity treatment in the average effect models (i.e., Eq. 1), we employed a distance decay version of the DID approach, as shown in Eq. 2. The distance decay study uses proximity intervals ($T_i^m,\forall m\le M-1$) as the treatment variables instead of a single binary treatment (as $T_i$ in Eq. 1). The distance-decay model shown in Fig. 2 uses 0.5-mile intervals from 0 to 6 miles, with properties in the 5 to 6 mile ring (i.e., $T_i^M$) serving as the control group. To investigate the role of visibility, we further interact the proximity intervals with a binary visibility variable to produce the results in Fig. 1 (i.e., the treatment variables become $T_i^m\times 1(View=1)$ and $T_i^m\times 1(View=0)$). The model specifications in Eq. 2 are identical to Eq. 1 except for differences in the treatment variables,

$$\begin{array}{c}\text{ln}\left(P_{\mathit{it}}\right)={\beta }_0+{\beta }_1{\mathit{Post}}_{\mathit{it}}+{\displaystyle \sum _{m=1}^{M-1}}{\beta }_2^m{T}_i^m+{\displaystyle \sum _{m=1}^{M-1}}{\beta }_3^m{\mathit{Post}}_{\mathit{it}}\times T_i^m\hfill \\ +{\delta }_k{\displaystyle \sum _{k=1}^K}{X}_{\mathit{it}}^k+{\gamma }_k{\displaystyle \sum _{k=1}^K}({\mathit{Post}}_{\mathit{it}}\times X_{\mathit{it}}^k)+{\tau }_{\mathit{ct}}+{\epsilon }_{\mathit{it}}.\hfill \end{array}$$ [2]

Based on the proximity cut-off point suggested in the distance decay results, we specify a proximity treatment (i.e., results suggest properties within 3 miles) for the average treatment model in Eq. 1. Moreover, we can test the average treatment effect of the interaction between visibility and proximity, by slightly modifying Eq. 1 to allow for two treatment groups [i.e., effects shown as ${\beta }_3^{\mathit{view}}$ and ${\beta }_3^{no\_view}$ in Table 1 column (2)]. The empirical results of these specifications decide the appropriate treatment to use for subsequent studies, where the control group specification will also be consistent with the exploratory specifications.^¶¶ Details of subsequent event study and heterogeneity analyses are presented in SI Appendix.

Our DID model relies on the assumption that the LSSPV siting process is independent of the price trends over time conditional on the covariates (i.e., the parallel trends assumption). We conduct pretrend tests with placebo treatments by setting a pseudo-post variable mimicking a fake installation event 6 y before the actual installation and dropping observations that are actually treated after the actual site installation. Null effect estimates from the placebo tests support the plausibility of the parallel trends assumption. Moreover, the event study model could also display pretreatment effects where pretreatment trend differences would show up and suggest a violation of the parallel trends assumption.

3.5.2. Analyses for agricultural land and large-lot homes.

We use a distance-decay model to detect the cut-off proximity for the treatment variable in the DID analysis for agricultural or vacant land and large-lot homes as the potential impact mechanism is related to site proximity. The ag-land distance-decay model is built on Eq. 2 with three key differences. First, the outcome variable is the natural logarithm of land price per acre. Second, the control variables $X_{\mathit{it}}^k$ do not include house characteristics. Finally, based on the volume of ag-land transactions, the proximity intervals are selected every two miles from 0 to 20 miles, the fixed effects used are on the county-site (i.e., an interaction between county and the LSSPV site identifier)^## by year level, and the SE are two-way clustered at the county-site and year level. We also conduct the event study and heterogeneity analyses using the treatment variable suggested by the ag-land distance-decay model. Furthermore, we conduct pretrend tests for the ag-land analysis to check the plausibility of parallel trends assumption. The large-lot-home analysis retains the outcome and control variables from the residential analysis while adopting the same proximity bins and fixed effects used in the ag-land analysis. More details of ag-land and large-lot-home analyses are provided in SI Appendix.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Our replication package (https://github.com/Starfallchen/SolarViewHedonic) provides all code used in this study, including Stata and Python code for raw data processing, geospatial variable processing, viewshed analysis, data aggregation, and estimation analysis (45). All analyses are conducted in Stata 18MP (https://www.stata.com/order/) (46) and Python 3.9.18 https://www.python.org/downloads/release/python-3918/) (47). The replication package also shares datasets that are from unrestricted data sources. The property transaction data are acquired from CoreLogic Solutions, LLC (https://www.corelogic.com/360-property-data) (48). Restricted by contract with CoreLogic, all variables derived from raw CoreLogic data will not be shared. To replicate our study, we recommend acquiring CoreLogic national-level property data with transactions from 1993 to 2020 and applying the data processing code in the replication package. Other raw data are from publicly available sources. The large-scale solar site data are available at the US Large Scale Solar Photovoltaic Database webpage: https://eerscmap.usgs.gov/uspvdb (49). Digital Elevation Models in the viewshed analysis are produced by NASA’s Shuttle Radar Topographic Mission and available at https://srtm.csi.cgiar.org (50). Geospatial data on states, counties, census tracts, primary roads, and metropolitan areas are from US Census Bureau TIGER/line geodatabase, available at https://www.census.gov/geographies/mappingfiles/time-series/geo/tiger-geodatabase-file.html (51). Geospatial data on transmission lines are obtained from US Energy Atlas hosted by Energy Information Administration, available at https://atlas.eia.gov/search (52). Data for heterogeneity analysis are drawn from multiple public sources, with details described in the SI Appendix.