Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Environmental Exposure Measurement and Place-Based Causal Inference

Abstract. Environmental Exposure Measurement and Place-Based Causal Inference examines how geographic, sensor-derived, activity-space, and experienced environmental measures can support defensible causal claims about psychological functioning. This fundamental critical narrative monograph reconstructs neighborhood research and environmental exposure science, defines the uncertain geographic context and modifiable areal unit problems, and analyzes spatial support, geocoding, remote sensing, air and noise measurements, residential histories, GPS, geographically informed ecological momentary assessment, street-view imagery, and perceived environments. It develops a causal framework covering potential outcomes, directed acyclic graphs, spatial selection, interference, target trials, natural experiments, staggered difference-in-differences, longitudinal g-methods, mediation, and sensitivity to unmeasured confounding. Primary empirical evidence includes randomized vacant-lot greening in Philadelphia, the Moving to Opportunity housing experiment, school greenness and cognitive development, cross-national transportation noise, co-occurring pollution and greenness, and residential relocation in Bradford, and recent ecological-context sensitivity research. The synthesis distinguishes causal intervention effects from longitudinal and cross-sectional associations, addresses environmental justice, measurement inequity, reproducibility, and transportability, and specifies research priorities for the Artificial Era. The research integrates identifiable sources without claiming an independently conducted systematic review or meta-analysis.

Keywords: environmental exposure measurement; place-based causal inference; environmental psychology; geographic information systems; GIS buffers; geocoding; spatial epidemiology; modifiable areal unit problem; uncertain geographic context; daily activity spaces; remote sensing; vegetation index; environmental justice; residential selection; causal diagrams; target trial emulation; natural experiments; difference-in-differences; randomized greening; moving to opportunity; psychological well-being; exposure misclassification; interference; longitudinal causal inference; sensitivity analysis; Artificial Era

The Scientific Problem of Measuring Place

Environmental exposure measurement and place-based causal inference study how environments become observable variables and how changes in those environments can be related to psychological outcomes under defensible causal assumptions. A setting contains physical conditions, social arrangements, institutional constraints, and personally experienced meanings. An exposure is a specified contact with, opportunity to encounter, or experience of an environmental attribute during a defined period. A contextual association is the observed relationship between such an attribute and an outcome; a causal effect concerns how the outcome would change under a specified alternative environmental intervention. A census tract, a park polygon, and a recorded mood score participate in different measurement processes. Their spatial coincidence cannot by itself establish the action of an environmental cause. Diez Roux (2001) Kwan (2012)

The explanatory unit is often a person moving among several places through a sequence of days and years. Residence may determine overnight air quality while commuting governs traffic noise, workplaces shape thermal exposure, and institutions determine access to restorative amenities. Exposure duration, intensity, route, opportunity, and subjective interpretation can interact. Psychological mechanisms include attentional restoration, stress appraisal, felt safety, perceived control, sleep disturbance, social interaction, and developmental learning. Each mechanism implies a distinctive exposure window, spatial geometry, and outcome schedule. A strong design therefore defines the substantive process before selecting the convenient geography or available sensor. Helbich (2018) Hartig et al. (2014)

This monograph develops a critical narrative scientific synthesis, moving from historical geography and environmental epidemiology through contemporary mobile sensing and causal identification. It examines representative original trials, natural experiments, longitudinal cohorts, and methodological studies, with their design-specific evidentiary limits. Source records are linked directly through identifiable citations; no new participant data have been collected, no formal systematic-search protocol has been executed, and no pooled meta-analysis is claimed. The central thesis is methodological: environmental exposure validity and causal identification are jointly necessary. More precise maps improve causal research only when those maps represent a meaningful intervention and are integrated with appropriate comparison groups, temporal ordering, and outcome measurement. Hernán & Robins (2016) Rohrer (2018)

From Environmental Context to Explicit Exposures

Environmental psychology historically considered how built, natural, and social settings organize experience, while epidemiology developed methods for comparing geographically patterned risk. Neighborhood health research inherited administrative data because addresses, census counts, and disease registers could be joined with relatively consistent boundaries. In a methodological overview, Diez Roux (2001) distinguished ecological, contextual, multilevel, and comparative neighborhood approaches and stressed the problem of separating compositional from contextual explanations. A neighborhood coefficient may reflect unmeasured household resources or life-course selection; its interpretation requires a model of why location matters. The history of the field is consequently a history of improving both environmental concepts and study designs. Diez Roux (2001) Chaix (2009)

Geographic information systems made it feasible to derive exposure indicators from modeled pollutant surfaces, vegetation indices, road networks, land uses, and distances to amenities. These methods expanded comparative scale and reproducibility. They also elevated implicit decisions about geographic extent, resolution, time, and feature classification into scientific assumptions. A small circle around a residence might capture immediate street vegetation, whereas an accessible network buffer might capture practical walking opportunities. Their difference can change which participants are assigned exposure and the substantive meaning of the regression coefficient. James and colleagues' review of greenness research illustrates the varied operationalization of natural environments and the importance of outcome and exposure specificity. James et al. (2015) Hystad et al. (2011)

A second transformation placed individual mobility and daily trajectories at the center of environmental assessment. Helbich (2018) argued that residential snapshots omit environments encountered along ordinary paths and across residential histories. Kwan (2012) formulated the uncertain geographic context problem, showing why ambiguous geographic boundaries and exposure timing generate inferential uncertainty beyond familiar changes in areal aggregation. Contemporary studies can combine address histories, sensor measurements, and timed behavior reports; those capabilities increase the obligation to examine who is tracked, which activities are missing, and whether the exposure proxy captures the psychological construct. A technical advance supplies a new source of evidence rather than immunity from selection and interpretation. Helbich (2018) Kwan (2012)

Construct Definition and the Intervention Question

The term environment can denote vegetative cover, accessible public space, ambient concentration, noise level, perceived danger, neighborhood cohesion, or an institutional resource. These are separate constructs, even when correlated. Vegetation cover estimated from satellite pixels is an ecological surface attribute. Park access is a relation between people, travel routes, schedules, restrictions, and services. Perceived safety is a psychological appraisal influenced by events, memories, group belonging, and public maintenance. The research question must state which of these is exposure, which is mediator, and which is the intended outcome. Otherwise a study can shift silently between land cover, recreation, and therapeutic restoration while retaining a single label such as green space. Hartig et al. (2014) Kondo et al. (2018)

A causal question can be expressed through potential outcomes: for a defined population, how would a prespecified psychological outcome differ if a city implemented an environmental change rather than an alternative policy? A vacant-lot greening program is a manipulable package; a satellite-derived greenness score is normally an indicator whose multiple possible interventions are not equivalent. Planting trees, opening public access, improving maintenance, and eliminating traffic can all raise perceived environmental quality by different routes. The exposure contrast therefore requires a treatment definition, outcome time, implementation mechanism, eligibility criteria, and intended population. Consistency becomes difficult when nominally identical exposure levels represent materially different lived experiences. Hernán & Robins (2016) South et al. (2018)

For observational research, investigators should formulate the hypothetical target experiment before conditioning on covariates. Specify who would qualify at baseline, the time zero, intervention strategies, follow-up, and primary estimand. In a longitudinal policy evaluation, a resident's moving away is neither automatically a control observation nor always a reason for exclusion: mobility may respond to the intervention itself. Similarly, controlling for postintervention perceived safety may remove a pathway through which greening affects distress. Defining the estimand first clarifies whether a variable should be measured as baseline confounder, mediator, moderator, competing exposure, or censoring process. Hernán & Robins (2016) Greenland et al. (1999)

Spatial Scale, Support, and the Modifiable Areal Unit Problem

Every spatial exposure is measured on a support: a pixel, point, buffer, street segment, building, block, tract, route, or jurisdiction. Statistical aggregation changes the distribution of values and can alter associations. The modifiable areal unit problem refers to sensitivity arising when observations are grouped into different zone systems or spatial scales. It matters for psychological research because average tract noise, block greenery, and the individual path of a child to school characterize different actual encounters. A study that uses areas defined for administrative convenience may be measuring a different exposure than its theory requires. Multiple geographic specifications should be understood as different operational estimands, not merely alternative settings of the same numerical algorithm. Kwan (2012) Diez Roux (2001)

Residential buffers are commonly defined by straight-line radius. A circular buffer includes places separated by rivers, restricted land, or highways; a street-network buffer traces possible travel along accessible routes; a pedestrian travel-time catchment adds slope, crossings, physical capability, and network friction. A density raster averages a surface, while a nearest-park metric privileges the first accessible facility regardless of quality. For noise and air pollution, modeled dispersion and vertical position may matter more than geometric proximity. Scale selection must correspond to environmental transport and behavior. Predefined sensitivity analyses can compare plausible supports, but searching many radii and selectively reporting the strongest association yields a multiple-testing problem. Hystad et al. (2011) James et al. (2015)

Geographic boundaries can also be politically and historically consequential. A single tract indicator compresses internal variation in residential blocks, school catchments, amenity access, and environmental disinvestment. Different populations may encounter the same geographic unit through distinct transportation, work, and caregiving routines. Across cities and countries, tract dimensions and institutional meanings vary, impairing comparability even when nominal measurements coincide. Consequently, a research protocol should document original data resolution, boundary vintage, geocoding precision, spatial joins, raster aggregation, service-area assumptions, and planned alternative supports. Such provenance allows another researcher to distinguish an error in mapping from a disagreement about contextual theory. Diez Roux (2001) Kwan (2012)

The Uncertain Geographic Context Problem

Kwan's uncertain geographic context problem concerns uncertainty about where and when a context influences a person. The boundary of an investigator-defined neighborhood rarely reproduces the true context of school attendance, interpersonal contact, inhalation, or emotional appraisal. The uncertainty is spatial because relevant settings extend differently around each person and temporal because they appear at particular times or accumulate over longer periods. It cannot be resolved simply by selecting finer administrative polygons: a precisely located home may remain a poor proxy for workplace exposure and everyday movement. The scientific question is therefore how environmental contact is generated in the first place. Kwan (2012)

Geographic context uncertainty differs from ordinary errors in the recorded value of an exposure. A perfectly calibrated air monitor at the correct coordinate can still represent the wrong location for a commuting worker. Likewise, detailed annual vegetation data can represent the wrong temporal window for an acute mood episode. GIS precision addresses the sensor-to-place relationship; inference needs the place-to-person and time-to-process relationships as well. The uncertainty can be reduced through diaries, mobile sampling, repeated measurements, and mechanism-specific windows, while each method introduces participant burden and privacy constraints. Triangulating measurement approaches is useful when the disagreement is itself substantively informative. Helbich (2018) Kwan (2018)

Neighborhood effect averaging further complicates residential exposure comparisons. Kwan (2018) described how people living in contrasting residential contexts may visit more similar nonresidential environments, thereby shrinking the difference in their total encountered exposures. If travel opportunities differ by socioeconomic resources, residential comparisons can generate complex distortions rather than a universal attenuation factor. A complete analysis may therefore estimate both the residential policy contrast and the actual activity-space contrast, interpreting them as distinct questions. Adjusting for postresidential mobility indiscriminately may remove a genuine pathway by which an environment changes behavior. Kwan (2018) Rohrer (2018)

Psychological Processes in Environmental Exposure

Exposure operates through physical and psychosocial pathways with different time constants. Noise can interrupt classroom information processing or sleep; heat can alter comfort and task persistence; green infrastructure may facilitate restorative experiences, outdoor activity, social contact, or reduced pollution. Hartig and colleagues (2014) organized major candidate pathways from nature to health through air quality, physical activity, social cohesion, and stress reduction. Their framework is explanatory: demonstrating that vegetation is correlated with well-being does not identify which pathway carries an effect. The hypotheses require direct measures and appropriate temporal ordering. Hartig et al. (2014) Stansfeld et al. (2005)

Appraisal mediates some environment–psychology relationships but is also a meaningful outcome in its own right. A resident may experience a maintained park as safe, socially welcoming, or culturally alienating; these appraisals affect whether nominal access becomes contact. Perceived disorder and fear may be altered by cleaning vacant sites, while objective crime counts reveal a partially independent outcome domain. Branas and colleagues' cluster randomized intervention linked urban vacant-land remediation to changes in violence and fear, motivating joint but distinguishable measurements of physical conditions, experienced safety, and social behavior. Branas et al. (2018) Kondo et al. (2018)

Psychological exposure histories also depend on sensitivity periods. Children's schooling, neurocognitive development, and later life opportunities unfold over years, while daily distress measures detect shorter changes. Dadvand and colleagues' longitudinal school study associated surrounding greenness with changes in cognitive development; its repeated measurements strengthen temporal description but observational covariates cannot eliminate every school-level or family-level alternative. Randomized housing-voucher research, by contrast, identifies effects of a policy offer for eligible families, while attributing those outcomes to a single neighborhood attribute remains difficult. Measurement and design must be judged jointly for each causal claim. Dadvand et al. (2015) Ludwig et al. (2012)

A Methodological Architecture for Evidence Synthesis

An evidence hierarchy for this subject must separate the scientific question from the design label. A cluster randomized greening trial can estimate the effect of an intervention at selected lots among sampled residents; it does not automatically estimate the effect of every type of greenness throughout an entire metropolitan area. A longitudinal residential cohort can describe within-person environmental change; it may remain vulnerable to time-varying shocks that induce both relocation and distress. A city-level natural experiment may have strong temporal contrast but weak information about actual individual contact. These sources can inform one another precisely because they answer different, adjacent estimands. South et al. (2018) Subiza-Pérez et al. (2025)

The synthesis developed here treats primary trials and methodological articles as anchors, then examines thematic reviews for coverage and alternative interpretations. Explicit comparative dimensions include exposure construct, spatial support, temporal window, unit of assignment, person-level selection, outcome measurement, identifying assumption, sensitivity analysis, and population transport. The principal threat is often a compound mismatch: a valid GIS measure used for an ill-defined intervention, combined with an observational estimator that silently presumes exchangeability. An impressive sample size cannot repair this mismatch. Hernán & Robins (2016) Helbich (2018)

Every claim of evidence strength is conditional on its source. The historical articles establish conceptual problems; original studies establish observed associations or estimated trial contrasts under their designs; reviews summarize a changing literature. Numerical estimates are presented only where the underlying publication gives the relevant denominator, design, and scale. Because the material integrates studies without a reproducible systematic screening and risk-of-bias protocol, it supports a critically argued research monograph rather than a systematic-review prevalence estimate or a new pooled causal effect. This distinction is part of scientific provenance, essential to ongoing revision as environmental measurement technologies and causal methods improve. Gascon et al. (2015) Diez Roux (2001)

Geocoded Addresses and Residential Histories

Residential addresses provide a tractable bridge between administrative records and spatial environmental data. Address matching assigns a coordinate or area identifier to a residence, after which investigators extract a pollutant estimate, land-cover value, street feature, or jurisdictional attribute. The procedure requires a reference date: an individual who moved between two addresses should not be assigned an exposure derived solely from the most recently recorded residence over an earlier outcome period. Geocoding can place addresses at building centroids, interpolated street segments, or postal-area centroids, each with different error structures. Research documentation should identify geocoder version, match quality, coordinate reference system, and geographic confidentiality transformations. Hystad et al. (2011) Helbich (2018)

A residential history is an ordered sequence of places with dates and intervals of residence. It permits questions about cumulative exposure, age-specific windows, residential transitions, and delayed outcomes. For instance, a study of adolescent development might calculate years lived within specified conditions rather than treat the current neighborhood as the full developmental environment. Chetty and colleagues (2016) used a randomized housing-mobility experiment with administrative outcome linkage to illuminate differences by age at move, a result whose interpretation depends explicitly on childhood exposure time. A residential chronology therefore improves conceptual alignment even when daily activity remains unavailable. Chetty et al. (2016)

Residence histories also carry differential missingness. People facing unstable housing, disaster displacement, informal accommodation, or frequent changes in household composition may be harder to follow than stable homeowners. If instability affects both the probability of a recorded address and psychological outcomes, complete-case restrictions can select a socially advantaged sample. Privacy-preserving linkage can preserve a derived exposure history while limiting retained precision, but suppressed coordinates reduce future reanalysis options. Studies should report unmatched address proportions, time without valid residence, and uncertainty from interpolation. Those errors need to be examined by age, income, and residential stability before interpreting place disparities. Diez Roux (2001) Chaix (2009)

Satellite Greenness, Land Cover, and Vegetation Indices

Remote sensing supplies spatially extensive, repeatable observation of vegetation and related environmental features. Normalized Difference Vegetation Index (NDVI) uses the contrast between near-infrared and red spectral reflectance to indicate vegetative activity; Enhanced Vegetation Index incorporates additional corrections and responds differently under dense vegetation conditions. Neither index directly observes willingness to use a park, visual contact with trees from a dwelling, accessibility, nighttime safety, or the quality of a public space. An exposure labeled greenness therefore needs a documented choice of sensor, pixel resolution, acquisition season, cloud handling, and spatial summarization. James et al. (2015) Werder et al. (2024)

The relevant ecological scale differs by outcome. Greenness immediately outside a classroom may relate to visual attention and microclimate, while a larger residential catchment may stand for nearby outdoor recreation. Dadvand and colleagues (2015) assessed surrounding greenness at homes, schools, and on commuting routes when studying cognitive development in Barcelona schoolchildren. Their design illustrates the scientific value of measuring multiple actual environments rather than substituting residential vegetation for children's entire ecological experience. It also illustrates a remaining challenge: correlated features of schools and families can be difficult to separate observationally. Dadvand et al. (2015)

Seasonality changes the exposure meaning. Vegetation scores differ across droughts, deciduous seasons, agricultural cycles, and redevelopment; averaging an annual raster can erase exposure around a psychologically salient event. Satellite data can also underrepresent tree canopy beneath cloud cover and distinguish vegetation poorly from other reflective surfaces in heterogeneous cities. Multi-year maps may be helpful for stability but should not be applied retrospectively without checking changes in the landscape. A research-quality exposure specification reports exactly what is observed, how often, at what resolution, and which lived process that observation plausibly indexes. James et al. (2015) Gascon et al. (2015) The systematic review by Bardhan and colleagues (2024) examined greenery and bluespace visible in street-view images and mental-health associations. It identified low or very low certainty for the examined mental-health outcomes and emphasized inconsistent processing methods and geographic representation, showing why street-level visibility must be validated as a distinct exposure rather than used as a straightforward replacement for satellite land cover. Bardhan et al. (2024)

Air Pollution Models and Personal Dose

Air-pollution assessment ranges from fixed monitoring stations and dispersion models to land-use regression, satellite-derived estimates, microenvironmental sensors, and direct personal samplers. Model outputs for PM2.5 or nitrogen dioxide typically represent an estimated ambient concentration over a grid and time interval. They are distinct from inhaled dose, which depends on indoor infiltration, time spent in traffic, physical exertion, ventilation, and physiological characteristics. Hystad and colleagues (2011) combined geographic information with pollution modeling to develop population exposure surfaces across Canada, demonstrating broad coverage alongside the need to validate predictions where monitoring data are available. Hystad et al. (2011)

Assigning an annual average ambient concentration at a residence is defensible for some questions about chronic surrounding pollution. It is a weaker measure of short-lived exposure peaks or person-specific absorption. Differences between modeled and personal exposure may also correlate with occupation, household conditions, transit mode, and urban form. A classical measurement-error model may predict attenuation under restrictive assumptions; real spatial misclassification can be systematic and correlated with confounders. Investigators should evaluate monitor calibration, cross-validation, spatial residual patterns, indoor/outdoor differences, and errors specific to populations living near roads or industrial facilities. Hystad et al. (2011) Chaix (2009)

Multiple pollutants often share emission sources, meteorological conditions, and socioeconomic geography. Werder and colleagues (2024), using residentially linked PM2.5, nitrogen dioxide, and vegetation estimates among participants in the U.S. Gulf Long-term Follow-up Study, observed associations with depression that depended on joint environmental specification. That cross-sectional comparison cannot by itself assign a causal effect to a single exposure, but it demonstrates why interpreting one pollutant's coefficient without examining correlated exposures is hazardous. Exposure mixture models should state whether the scientific estimand concerns a single separable intervention, a combined environmental strategy, or a predictive profile. Werder et al. (2024)

Noise, Heat, Light, and Temporal Resolution

Environmental sound can be represented by modeled traffic contours, schoolyard measurements, or wearable acoustic observations. Average sound level alone may obscure peaks, timing, predictability, frequency, and the distinction between disruptive noise and desirable sound. Stansfeld and colleagues (2005) compared aircraft and road traffic noise exposures with children's cognitive outcomes in schools across three European countries. The study illustrates multilevel exposure measurement that combines maps with on-site environmental data, while its cross-sectional structure requires care in distinguishing environmental effects from residual differences between schools and surrounding areas. Stansfeld et al. (2005)

Thermal conditions combine ambient weather with building envelope, shade, ventilation, surface material, urban heat islands, and activity level. Lighting depends on intensity, spectrum, timing, glare, and availability of daylight. Exposure during a sleep period need not be equivalent to a daytime weekly average, even where a municipality-level index is identical. Psychologically meaningful windows may correspond to interrupted instruction, bedtime, emergency heat periods, or perceived discomfort during commuting. Studies should prespecify temporal aggregation and connect it to the proposed mechanism rather than assuming the most convenient weather or noise summary captures all relevant exposure. Helbich (2018)

Coincident stressors can create compound conditions: traffic corridors involve pollution and noise, dense built form shapes heat and access to vegetation, and working conditions govern both material exposure and perceived control. A model with many correlated environmental indicators may suffer weak positivity and unstable coefficients when no participants experience the hypothetical combinations being compared. Policy packages such as traffic calming or building rehabilitation often alter several dimensions together. The most interpretable estimator may consequently target a defined package rather than claim a precisely separable independent effect for each index. Hernán & Robins (2016) Greenland et al. (1999)

Daily Activity Spaces, GPS, and Ecological Momentary Assessment

GPS traces, transportation records, and time-use diaries can approximate where a person traveled and how long particular environmental features were encountered. An activity-space exposure might average pollutant concentrations along a route weighted by dwell time, or classify time spent inside and outside green environments. It is necessarily a modeled quantity because satellite location error, missing indoor signals, sensor nonwear, and sparse temporal sampling affect the observed path. Mobile assessment also changes participant behavior and can select people who own compatible devices or agree to carry sensors. Kwan (2012) Helbich (2018)

Ecological momentary assessment (EMA) asks people to report mood, stress, attention, or perceived safety in temporal proximity to daily experiences. Paired with movement histories, it can distinguish immediate within-person associations from static between-person neighborhood contrasts. Repeated reports may establish that a distress increase followed a particular encounter, yet that sequence alone does not remove time-varying confounding: activity choice can be affected by prior mood, obligations, weather, or social support. More severe symptoms may themselves reduce willingness to travel into parks, producing reverse directionality when visit counts and mood are regressed against each other. Rohrer (2018) Helbich (2018) A systematic review of geographically informed EMA by Christensen and colleagues (2025) identified 33 eligible studies and noted that longitudinal causal temporal analyses remained scarce despite the method's ability to record momentary place experiences. This review supports the measurement potential of GEMA while separating real-time association from causal identification. Christensen et al. (2025)

Exposure-weighted trajectories need a scientifically specified dwell-time or dose function. A ten-minute transit peak is not always commensurate with an hour of low-level home exposure, and an algorithm that interpolates between sparse fixes may cross places never actually visited. Privacy risk is also unusually high: repeated routes can reveal homes, workplaces, health care visits, and interpersonal routines. Strong research governance requires informed consent, minimization of collected precision, separate key management, restricted analysis, and disclosure of the uncertainties introduced by spatial masking. Collection of more coordinates is justified by a measurable improvement in the research estimand, not by their mere availability. Kwan (2018)

Perceived Environment, Place Meaning, and Construct Validity

An objective environmental indicator and a subjective appraisal measure different aspects of person–environment transactions. Observed greenery, perceived restorativeness, trust in local safety, and actual park visitation may all diverge. A park with mature trees may feel inaccessible if crossing roads is difficult or harassment is feared. A seemingly unattractive vacant area may carry strong social significance for nearby residents. Psychometric questions therefore arise: Does a safety item have similar meaning across neighborhoods? Do respondents use comparable reference periods? Does an appraisal scale distinguish momentary emotion from stable place attachment? Hartig et al. (2014) Kondo et al. (2018)

Subjective exposure is sometimes an intermediate mechanism. If greening reduces perceived threat, adjustment for perceived threat in the total-effect model can obscure the intervention's psychological influence. In other contexts, appraisal may be a baseline cause of residential selection and must be measured before treatment. A useful causal diagram distinguishes prior affective functioning, neighborhood choice, actual built conditions, experienced contact, perceptual appraisal, and subsequent outcomes. A longitudinal survey that asks about prior conditions retrospectively can suffer recall effects correlated with current distress, especially when environmental conditions are emotionally salient. Greenland et al. (1999) Rohrer (2018)

Construct validation should use convergent and discriminant evidence, cross-group measurement checks, and multimethod assessment. Agreement between satellite canopy and participant-rated greenness may support some constructs while disagreement can demonstrate that accessibility or perceived quality is missing from the geographic proxy. Researchers should resist treating an objective sensor as universally superior to psychological report: the proper indicator is the one measuring the causal mechanism of interest. For sensory exposure, devices may be essential; for fear, belonging, or perceived control, experience is itself part of the outcome domain. Diez Roux (2001) Hartig et al. (2014)

Exposure Error, Calibration, and Triangulation

Spatial measurements can deviate from the underlying target through geocoding error, temporal mismatch, instrument noise, model extrapolation, ecological aggregation, imprecise activity schedules, and construct substitution. These mechanisms require different corrections. Recalibrating a sensor cannot repair assigning the wrong neighborhood, and using a more detailed raster cannot repair an unmeasured migration history. An explicit measurement model links the observed value, target exposure, covariates, and error-generating process, and may require validation subsamples with personal monitoring or direct observation. Hystad et al. (2011) Kwan (2012)

Validation should examine both prediction accuracy and causal relevance. A land-use model may predict ambient pollutant concentration well at held-out stations but correlate weakly with individual indoor dose. GPS interpolation may reproduce recorded stop locations yet omit salient school settings. A self-report may demonstrate acceptable internal consistency while reflecting the outcome itself through negative affect. Repeated measures, blinded environmental audits, calibration studies, sensor co-location, and comparison of alternative spatial supports help characterize these weaknesses. The purpose is to identify which claims remain credible after accounting for plausible error magnitudes. James et al. (2015) Rohrer (2018)

Triangulation combines designs or measurements whose biases are sufficiently different to yield informative convergence. Agreement among address-based, daily-route, and objective park-access indicators may strengthen an interpretation, but their errors can be correlated by common mapping sources or selection into the sampled population. Disagreement is informative when it points to a mismatch between opportunity and contact or between physical surroundings and psychological appraisal. Reports should preserve alternative specifications and justify the final estimand, rather than treating one preferred statistically significant measure as the canonical exposure after the fact. Kwan (2018) Helbich (2018)

Causal Graphs for Place-Based Studies

A directed acyclic graph represents assumptions about causal relationships among exposure, outcomes, confounders, mediators, and selection. For place-based psychology, a simplified graph may connect parental resources to neighborhood selection and children's learning, place characteristics to school and stress exposure, and preexisting health to mobility. This makes clear why adjusting for observed socioeconomic variables can reduce some confounding without proving that unmeasured family expectations, employment constraints, or discrimination have been removed. The graph is a scientific model whose assumptions must be argued from substantive knowledge, not generated from the observed statistical association alone. Greenland et al. (1999) Rohrer (2018)

Time ordering is decisive. Prior income can confound the link between green neighborhoods and depression, while income changes caused by an environmental policy may be part of the causal pathway. Perceived neighborhood safety measured after greening can mediate psychological benefit; the same scale measured before a relocation may predict both destination choice and future distress. Controlling for a variable influenced by both an exposure and an unmeasured cause of outcome can introduce collider bias. Recording the same constructs at distinct dates is therefore a design requirement, not a cosmetic refinement. Pearl (1995) Hernán & Robins (2016)

Spatial dependence complicates graph specification because an individual's environment may affect neighbors and a city's policies may influence several nearby areas. Traffic diversion, increased rents, relocation, and social spillovers create cross-unit pathways. A design in which control neighborhoods receive displaced traffic cannot use a simple no-interference interpretation without qualification. Graphs should distinguish assignment, uptake, actual contact, intermediate behavior, and exposure of adjacent units; maps should identify the geographic reach of plausible spillovers. Such documentation makes it possible to explain what a counterfactual comparison includes and which effects it cannot separate. Rohrer (2018) South et al. (2018) Reich and colleagues' methodological review (2021) treats spatial correlation and interference as central features of environmental causal inference, emphasizing that geographic proximity can change both identification and uncertainty. Reich et al. (2021)

Residential Selection and Spatial Confounding

Residential selection occurs because households choose or are constrained into locations partly according to income, employment, caregiving, cultural affiliation, discrimination, health, and information. These factors can affect psychological outcomes independently of the chosen location. Even when income, education, and demographic variables are measured, unobserved preferences or prior symptoms can remain correlated with the exposure. This is the central reason a comparison between residents of greener and less green neighborhoods is not automatically a causal effect of greenness. Contextual models may describe disparities usefully while leaving the counterfactual relocation or redesign question open. Diez Roux (2001) Ludwig et al. (2012)

Propensity score methods estimate the probability of observed exposure conditional on specified baseline variables, allowing weighting, stratification, or matching between measured-comparable participants. Rosenbaum and Rubin (1983) formalized their role in observational studies. In environmental research, propensity scores require clear exposure alternatives and common support; they cannot invent observations from affluent leafy neighborhoods for otherwise similar households living near industrial facilities when no overlap exists. Researchers should inspect covariate balance and the distribution of weights and report populations excluded by lack of positivity. Rosenbaum & Rubin (1983) Hernán & Robins (2016)

Residential moves may appear to remove person-level confounding in before–after contrasts, but the reason for a move can itself be causally important. Job loss, bereavement, illness, school changes, and household dissolution may alter both neighborhood quality and mental health. Within-person estimates control only stable unmeasured attributes, and they remain vulnerable to time-varying causes. Residential trajectories can also be affected by anticipation: a person who plans a move might experience stress or relief before the address changes. A well-defined move-based study therefore measures pretrend outcomes, move motivations, major concurrent events, and the time from change to outcome. Subiza-Pérez et al. (2025) Helbich (2018)

Potential Outcomes, Estimands, and Exposure Versions

A potential-outcomes analysis asks what would have happened under alternative assignments or strategies. An intention-to-treat estimand for a greening trial compares assignment to greening with assignment to an alternative among eligible clusters. A treatment-on-the-treated comparison addresses actual uptake and depends on further assumptions about compliance and interference. An effect of moving into a lower-poverty area refers to a housing intervention that changes many features simultaneously; it does not identify a universal coefficient for tract poverty alone. Precisely naming the intervention protects the analysis from confusing measured context with a manipulable causal treatment. South et al. (2018) Ludwig et al. (2012)

Exposure versions become problematic when a continuous score can be generated by disparate mechanisms. Two blocks with the same vegetation index may differ in shade, openness, safety, biodiversity, and accessibility. A hypothetical policy that increases NDVI by a fixed amount is therefore underdefined unless the mechanism is specified. For an inhaled pollutant, an intervention lowering ambient concentration via emissions reduction might also alter traffic, employment, and noise. Policy packages can be valid treatments if their components are described; causal interpretation then belongs to the package rather than to the isolated numerical proxy. Hystad et al. (2011) Hartig et al. (2014)

A causal estimand also identifies an outcome scale and population. Mean changes in distress scores, risk of crossing a symptom threshold, academic test scores, and self-reported well-being are not interchangeable. Differential effects across childhood, adulthood, and socially excluded groups cannot be summarized by a population average without examining heterogeneity. The target trial framework requires a common baseline, eligibility, follow-up, strategies, and outcome definition, with protocols designed to make comparisons interpretable. When a proposed urban policy has staggered implementation, group-specific effects should be aligned to their correct periods rather than averaged blindly. Hernán & Robins (2016) Callaway & Sant’Anna (2021)

Natural Experiments and Intervention Assignment

Natural experiments exploit changes such as park construction, transit extensions, emission restrictions, housing allocation, infrastructure repair, or school rezoning that occur independently of some individual choices. Their causal strength depends on the assignment mechanism and the availability of credible comparisons. A municipal decision may be targeted toward neighborhoods where distress or crime is already changing, which would invalidate an assumption that treatment timing is as good as random. Projects also differ in anticipation, implementation quality, maintenance, population displacement, and interactions with other reforms. The exposure map must distinguish planned policy, actual installation, functioning infrastructure, and individual contact. Hernán & Robins (2016) Branas et al. (2018)

Randomized clusters provide a more transparent assignment mechanism when feasible. South and colleagues (2018) randomized geographically grouped vacant lots in Philadelphia to greening, minimal cleanup, or control conditions. Their trial directly addresses effects of specified place interventions on sampled residents' self-reported mental health, while geographic spillovers, survey attrition, and context-dependent implementation remain matters for interpretation. A corresponding study by Branas and colleagues (2018) examined violence, crime, and fear after vacant-land restoration. The difference between these outcomes illustrates the importance of matching the inference to the psychological construct rather than inferring a universal environmental benefit from one endpoint. South et al. (2018) Branas et al. (2018)

A well-conducted quasi-experiment documents why intervention and comparison sites had similar untreated trajectories, what could induce simultaneous changes, and whether spillovers cross the comparison boundary. Field interventions face ethical and practical constraints: improvements may spread gradually, residents may move, and neighborhoods may be affected by regional shocks. A study may validly report a policy effect on sampled areas without proving a mechanism. Mechanism studies can add measures of fear, park use, social contact, and experienced control, but decomposition of treatment pathways generally requires assumptions beyond those supporting the total intervention effect. Imai et al. (2010) South et al. (2018)

Difference-in-Differences and Staggered Rollouts

Difference-in-differences compares outcome changes in exposed and comparison populations. Its key identifying condition is a parallel-trends assumption for untreated potential outcomes, possibly conditional on measured baseline information. A city that installs trees first in deteriorating streets may violate this condition if those streets would have had different distress trajectories even without planting. Graphical pretrend assessment helps describe plausibility but statistically nonsignificant pretrend tests do not demonstrate the assumption. Appropriate comparison groups, implementation dates, and outcome measurement periods must be fixed before inspecting the estimated treatment effect. Callaway & Sant’Anna (2021)

Modern staggered-adoption methods address heterogeneous timing and effect variation. Callaway and Sant'Anna (2021) define group-time average treatment effects and methods to aggregate them; Sun and Abraham (2021) show how conventional two-way fixed-effects event-study coefficients can be contaminated when treatment timing and effects differ. In an urban renewal program, an average that uses already-treated neighborhoods as controls can misrepresent the actual dynamic policy effect. Researchers should identify cohorts by rollout date, use not-yet-treated or never-treated comparisons where supported, and report heterogeneity across time since intervention. Callaway & Sant’Anna (2021) Sun & Abraham (2021)

Place-based designs must also account for spatially correlated disturbances. Residents in nearby blocks may share crime changes, heat waves, transportation strikes, or local economic shocks, so an analysis treating thousands of adjacent individuals as independent experiments understates uncertainty. Assignment may occur at the lot cluster, school, or municipality rather than the participant level. The correct level for inference should reflect assignment and residual spatial correlation. Spillovers and displacement can make an untreated block a partially treated comparison, motivating alternative buffer exclusion, spatial lag analysis, or explicit network-based estimands, with their identifying limits disclosed. Branas et al. (2018) South et al. (2018)

Time-Varying Confounding and Longitudinal Strategies

Repeated environmental exposure can change psychological functioning, which in turn influences subsequent exposure. For example, depression may change employment and daily mobility; an earlier neighborhood improvement may alter perceived safety and future time spent outside. A time-varying variable can then be both a mediator of earlier exposure and a confounder of a later exposure–outcome relationship. Conventional regression that adjusts for every updated variable can distort the total effect of an exposure history by conditioning on consequences of prior treatment. Robins's work on sustained exposures established a general framework for addressing such problems through g-methods under explicit assumptions. Robins (1986)

G-computation, inverse probability weighting for marginal structural models, and related approaches can estimate contrasts between specified dynamic strategies when sequential exchangeability, positivity, consistency, and adequate measurement hold. In an environmental policy study, inverse weights may balance measured predictors of each period's exposure assignment and dropout. Extreme weights can reveal insufficient overlap and generate fragile estimates. These methods are most useful when interventions can be described over multiple times, rather than when investigators have only a single current residential indicator with unknown historical meaning. Robins (1986) Hernán & Robins (2016)

Longitudinal administrative data can support exposure histories and repeated outcomes, but retrospective environmental assignment remains sensitive to changing geocoding coverage, inconsistent measurement scales, and migration. Subiza-Pérez and colleagues' Bradford relocation study used residence-linked green and pollution indicators with prescription outcomes and measured covariates. Its observed associations provide a substantive example of dynamic exposure research; relocation and prescription records are still subject to selection and outcome-proxy limitations. The design's appropriate scientific contribution is a temporally structured association with a causal interpretation conditional on untested assumptions, rather than a randomized environmental dose-response law. Subiza-Pérez et al. (2025)

Mediation, Spillovers, and Interference

Environmental changes can affect psychological well-being through several pathways simultaneously. Greening may reduce visual disorder, influence physical activity, alter social interaction, and change perceived safety; housing mobility can modify schools, peers, household finances, and stress. Statistical mediation requires assumptions about treatment–mediator, mediator–outcome, and posttreatment confounding. Imai and colleagues (2010) formalized approaches to causal mediation analysis, making explicit why a significant association between mediator and outcome does not establish a mechanism. A credible mediation claim needs adequate timing and a defensible model of pathways. Imai et al. (2010) Hartig et al. (2014)

Spillovers are substantive in place-based interventions. A park may serve residents beyond a predefined radius; pollution reductions move with atmospheric conditions; traffic redirected from a treated street can raise exposure elsewhere. Interference can change the treatment definition from a local individual intervention into a regional policy with a network of affected units. If spillovers are ignored, both treatment and control exposure may be misclassified and estimated local effects may fail to describe total system consequences. Environmental researchers should model realistic service catchments and include plausible interference zones in sensitivity analyses. Kwan (2012) Branas et al. (2018)

Displacement raises a distinct estimand question. A renewal initiative that benefits remaining residents may also increase rents and induce moves by vulnerable households. A study based only on those remaining after treatment can therefore estimate a selected effect rather than the effect on the initially eligible population. Similarly, improved access to green space may increase visitors from neighboring areas while changing the neighborhood's social experience. Evaluation should retain baseline cohorts where feasible, measure migration, and report both place-level outcomes and resident-level follow-up. The analytic unit should follow the ethical and scientific question concerning whose outcomes count. Diez Roux (2001) Ludwig et al. (2012)

Randomized Greening and Community Psychological Outcomes

The Philadelphia vacant-land program provides unusually direct evidence because treatment assignment preceded outcomes and targeted physical features of places. South and colleagues (2018) implemented a cluster randomized trial with 110 vacant-lot clusters assigned to greening, minimal cleanup, or no intervention. The intervention involved trash removal, grading, grass planting, some trees, fencing, and maintenance. The investigators surveyed community-dwelling adults before and after intervention and analyzed psychological distress components. The published primary report describes reductions in feeling depressed and feeling worthless for residents near greened lots relative to controls. The intervention bundle and the neighborhood context define the causal question that the trial can answer. South et al. (2018)

A causal interpretation of randomized assignment is strongest for the effect of offering the place-level intervention as implemented. It requires attention to survey attrition, clustering, possible spillovers, and the representativeness of sampled sites. Randomization of lots does not automatically make every participant's actual green-space contact random, because individuals choose their routes and activities after treatment. Importantly, the trial did not isolate the spectral color of vegetation, the physical cleaning, maintenance routines, visible social care, or changed expectations as separate randomized components. These pathways remain hypotheses requiring dedicated measurement. South et al. (2018) Hartig et al. (2014)

Branas and colleagues (2018) evaluated restoration of blighted vacant land and outcomes including violence, crime, and fear using a related citywide cluster randomized design. Their study connects an environmental intervention to measured public-safety outcomes. From a psychological perspective, lower fear and altered routine activity may contribute to changes in experienced place, but crime statistics and self-reported distress are distinct endpoints. The two trials show how community intervention studies can produce converging evidence across multiple outcome domains while preserving different causal estimands and measurement procedures. Their findings should motivate context-specific replication rather than an automatic claim that any urban planting program has the same effect. Branas et al. (2018) South et al. (2018)

Housing Mobility Experiments and Environmental Bundles

The U.S. Moving to Opportunity experiment offered randomly selected low-income public-housing families access to vouchers that facilitated relocation to less disadvantaged neighborhoods. Unlike observational comparisons of households already residing in different districts, randomized offers establish an identifiable policy contrast before residential decisions unfold. Ludwig and colleagues (2012) linked this design to long-run adult well-being and documented improvements in adult physical health, mental health, and subjective well-being after movement toward lower-poverty neighborhoods over a follow-up of roughly ten to fifteen years. Their estimates address a social and residential policy intervention, with assignment, uptake, and destination neighborhood characteristics that must be kept analytically distinct. Ludwig et al. (2012)

MTO demonstrates both causal leverage and interpretive complexity. Relocating households may experience changed safety, social relations, schooling, transportation costs, housing quality, and neighborhood services. A voucher offer can have a measurable effect without determining which element of that package produced it. Moreover, a treatment effect among families eligible for a historical housing demonstration is not necessarily the effect among all households contemplating a move today. Selection into voucher take-up and differing implementation pathways distinguish intention-to-treat effects from effects among actual movers. Attempts to identify those later contrasts require additional assumptions beyond initial randomization. Ludwig et al. (2012) Hernán & Robins (2016)

Chetty, Hendren, and Katz (2016) examined children's long-term outcomes using administrative data linked to the experimental housing intervention. Their results indicated more favorable adult education and earnings for children who moved at younger ages, with different patterns for adolescent movers. This design emphasizes timing and duration of environmental exposure, a scientific question invisible in simple cross-sectional maps. The finding should not be recast as evidence that a particular vegetation level or pollution index caused earnings changes. It instead supports the broader principle that histories of place-related opportunity can have age-dependent consequences whose component mechanisms require separate investigation. Chetty et al. (2016)

School Greenness and Cognitive Development

The study by Dadvand and colleagues (2015) investigated surrounding greenness and cognitive development among primary schoolchildren in Barcelona using repeated computerized cognitive assessments and geographically characterized exposure. It compared vegetation at homes, schools, and commuting routes, thereby making spatial context more informative than a single residential measure. Greater school greenness was associated with more favorable changes in selected cognitive measures over the period of observation. The design demonstrates how repeated outcome testing and multiple spatial supports can improve an environmental developmental analysis. It remains observational, so school resources, neighborhood conditions, and family-related selection must remain part of the causal appraisal. Dadvand et al. (2015)

Green-space exposure can operate through several candidate pathways: visual restoration, attention shifts, physical activity, peer interaction, lower air pollution, and altered noise or heat. A vegetation index cannot directly distinguish these mechanisms. Where greenness differs systematically between schools, a model must consider educational organization, classroom practices, school socioeconomic composition, traffic, and physical facilities. Mediator adjustment should follow a causal hypothesis instead of automatic inclusion of variables such as pollution that may themselves change following a greening intervention. The school is also a shared unit of exposure, creating clustering and correlated outcome trajectories among students. Dadvand et al. (2015) Hartig et al. (2014)

The contrast between the Barcelona school study and Philadelphia randomized greening is analytically productive. One provides repeated cognitive outcomes alongside observational geographic variation; the other supplies randomized alteration of neighborhood land conditions with psychological symptom reporting. Agreement on the plausibility of environmental benefit does not make their treatments, exposure windows, outcome constructs, or populations interchangeable. A future trial on school greening would need to define whether its target is the planting intervention, classroom views, usable outdoor environments, or reduced traffic, with blinded or standardized outcome assessment where feasible. Evidence synthesis should preserve those distinctions. Dadvand et al. (2015) South et al. (2018)

Traffic Noise and Learning Outcomes

The cross-national RANCH investigation by Stansfeld and colleagues (2005) studied aircraft and road traffic noise in relation to children's cognition and health. School noise was estimated through modeling and on-site measurements across sites near major airports in the Netherlands, Spain, and the United Kingdom. The published report assessed 2,844 children aged approximately nine to ten years. The exposure was materially and temporally specific to schools, with a clear educational relevance: recurring noise may interrupt comprehension, attention, and classroom learning. The multicountry framework strengthened the breadth of observation while retaining the limitations of a cross-sectional environmental comparison. Stansfeld et al. (2005)

Noise metrics pose particular measurement questions. Annual averaged exposure may conceal exam-period interruptions, peak aircraft events, and the timing of sound during lessons. Sound energy, annoyance, and perceived uncontrollability are related but distinguishable quantities. A school with high external noise may mitigate indoor exposure through insulation and building layout. Thus environmental noise measurements can be valid as source indicators while imperfect as measures of cognitive interruption experienced at the desk. Validity improves when measurement is aligned to the hours, rooms, and tasks relevant to learning. Stansfeld et al. (2005)

Causal interpretation requires attention to school selection, residential location, social resources, baseline educational performance, and differences across airports or municipal authorities. Even when schools are matched on measured socioeconomic attributes, unmeasured teaching conditions and family factors can remain. A policy that reduces aircraft noise, modifies insulation, or reschedules flights offers a more clearly manipulable contrast. Its effect on learning may depend on existing mitigation and the stability of school cohorts. Hence noise research illustrates the dual necessity of physical exposure calibration and careful comparison design. Stansfeld et al. (2005) Hernán & Robins (2016)

Air Pollution, Green Space, and Exposure Mixtures

Werder and colleagues (2024) examined co-occurring pollution and vegetation metrics in the U.S. Gulf Long-term Follow-up Study. Among 9,015 participants, residential addresses were linked to annual fine particulate matter, nitrogen dioxide, and satellite vegetation measures, and respondents completed standardized depression and anxiety questionnaires. The highest PM2.5 quartile was associated with a greater prevalence of depression, while higher greenness was associated with lower prevalence in mutually adjusted models. The published abstract reports a depression prevalence ratio of 1.17 (95% confidence interval 1.06–1.29) for the highest PM2.5 quartile and 0.89 (0.80–0.99) for highest greenness. These are adjusted cross-sectional associations, not randomized risk reductions. Werder et al. (2024)

The findings show why exposures should be analyzed jointly and specified transparently. Greener areas may differ systematically in traffic, noise, housing, income, neighborhood disadvantage, and opportunity for recreation. High collinearity among environmental conditions can make the independent effect of one marker difficult to interpret. Observed modification across social groups or urbanicity can motivate mechanistic questions and equity assessment, while multiple subgroup analyses create additional uncertainty. The proper synthesis reports model specification and study design alongside magnitude and confidence interval, without converting a prevalence ratio into individual therapeutic advice or a universal causal estimate. Werder et al. (2024)

Earlier reviews provide important context for interpretation. Gascon and colleagues (2015) systematically examined long-term residential green and blue space exposure and mental health; their review found insufficient or limited causal evidence for several exposure–outcome contrasts because definitions and designs varied. Hartig and colleagues (2014) synthesized plausible pathways and identified methodological gaps. The difference between an association and an intervention effect remains central even as geographic datasets become larger and satellite resolution becomes finer. Multiple measurements of correlated landscapes do not amount to independent causal confirmations unless their error structures and assignment conditions support the inference. Gascon et al. (2015) Hartig et al. (2014)

Residential Relocation and Time-Varying Exposure Evidence

Subiza-Pérez and colleagues (2025) studied within-city movers in Bradford, United Kingdom, using linked residential environmental indicators and health-record prescription outcomes. The analysis differentiated people with and without existing antidepressant or anxiolytic prescriptions at baseline and estimated associations between changes in greenness, particulate pollution, and subsequent recorded prescriptions. In the group without baseline prescriptions, greater residential NDVI changes were associated with lower odds of having an active prescription at follow-up, whereas increases in particulate concentrations were associated with higher odds. The paper's detailed mobility analysis provides valuable temporal structure while remaining observational. Subiza-Pérez et al. (2025)

Interpretation depends strongly on the outcome definition. An active prescription is an administrative measure of treatment and health-care interaction, not a direct diagnostic evaluation of depression or anxiety. Prescribing patterns can vary by access, clinician practice, need, and continuity of care. Relocation also may follow unobserved changes in employment, housing stress, or health. The authors' use of directed-graph-informed adjustment helps formalize assumptions but cannot guarantee that all confounders were observed. A modest exposure change estimated from a residential buffer may also imperfectly capture changes in overall daily contact. Subiza-Pérez et al. (2025) Rohrer (2018)

The Bradford setting demonstrates heterogeneity of opportunity: locations with greater vegetation are not necessarily safer or more usable, and the relation between greener and less deprived places can reverse across urban contexts. Generalizing from the study to a different country requires considering housing markets, park quality, health-care systems, baseline prevalence, and environmental geography. It is particularly informative when placed beside randomized greening and housing-voucher studies, because all three investigate place changes but differ in intervention control, what was measured, follow-up timing, and what counted as a psychological outcome. Subiza-Pérez et al. (2025) South et al. (2018) Ludwig et al. (2012)

Convergence, Contradictions, and Evidence Strength

The reviewed empirical literature supports different levels of inference. Randomized place interventions can support causal claims about implemented policy packages in the original trial populations. Randomized housing offers support causal claims about offers and, with stronger assumptions, uptake-related outcomes. Longitudinal school and relocation observations improve temporal analysis but do not establish exchangeability merely by including repeated measurements. Cross-sectional multi-exposure studies demonstrate associations and hypothesis-generating heterogeneity. These distinctions are fundamental to evidence synthesis and should remain visible when environmental metrics and outcomes are presented together. South et al. (2018) Ludwig et al. (2012) Dadvand et al. (2015) Werder et al. (2024)

Apparent contradictions can have substantive explanations. Planting a maintained vacant lot changes visual order and social signals; NDVI around a home may rise because of inaccessible roadside vegetation. Pollution measurements at residence may improve while commute exposures worsen. A before–after depression score may respond to an intervention during a short window, whereas an adult developmental trajectory requires longer follow-up. Differences in mechanisms, interventions, measurement, and populations can therefore produce nonidentical results without demonstrating that one design is necessarily defective. A critical comparison needs a matrix of exposures, estimands, and evidence quality rather than a simplistic count of positive studies. Helbich (2018) Kwan (2018) Gascon et al. (2015) A particularly recent example is Song and Luan (2026), whose analysis of 2018 Washington State death certificates found that estimated associations between residential NDVI and mental-illness-attributed mortality changed across ecological subregions and vegetation thresholds. Their analytic population comprised 31,245 decedents with 233 deaths in the specified mental-illness categories. Conditioning on decedent status, observational matching assumptions, and the relatively small outcome subgroup limit what can be inferred about the population-wide causal effect of greenness on mortality. The study illustrates the need for context-sensitive exposure definitions without resolving causal mechanisms. Song & Luan (2026)

Convergent claims are strongest when different biases are constrained and predictions are specific. A randomized greening trial can test the total local intervention effect; a mobile exposure study can test whether treated residents actually encountered the altered space; a longitudinal psychological survey can examine whether changes precede the hypothesized process. If intervention assignment, contact, mediator, and outcome are all observed, the resulting account can move beyond a single geographically aggregated association. Future studies should retain null and adverse results, preregister primary outcomes, release reproducible geographic processing code where privacy permits, and document mechanisms of attrition. South et al. (2018) Hernán & Robins (2016) Gascon et al. (2015)

Sensitivity to Unmeasured Confounding

A causal estimate derived from observational environmental data must be read in relation to the assumptions that identify it. Adjustment for documented income, age, education, and housing conditions addresses only the measured portion of residential selection. An unobserved change in employment, family care obligations, baseline distress, or anticipated redevelopment may influence both environmental exposure and subsequent well-being. Sensitivity analysis asks how strong such omitted relationships would need to be to change an inference. Cinelli and Hazlett (2020) express omitted-variable bias in terms of partial associations with treatment and outcome, enabling explicit benchmarks against observed covariates. Their approach is most informative when the analyst states a meaningful unmeasured scenario rather than converting a single sensitivity number into a proof of causality. Cinelli & Hazlett (2020) Rohrer (2018)

The E-value introduced by VanderWeele and Ding (2017) characterizes, on a risk-ratio scale under its assumptions, the minimum strength of association that an unmeasured confounder would need with both exposure and outcome to explain away an observed association. Such a calculation can provide a transparent robustness description for eligible effect estimates; it cannot test the absence of bias, settle measurement error, or ensure that causal ordering has been correctly specified. A result robust to one hypothetical confounder might still be distorted by spatial misclassification, selective migration, outcome ascertainment, or an improperly defined intervention. Reporting the chosen sensitivity parameter, effect scale, and interpretive boundary is therefore indispensable. VanderWeele & Ding (2017) Greenland et al. (1999)

Environmental research benefits from layered robustness tests. Investigators can compare alternative exposure supports chosen in advance, evaluate negative-control outcomes or exposures when their identifying assumptions are plausible, restrict analyses to locations with reliable geocoding, and assess overlap across socioeconomic strata. Within-person analyses can reduce bias from stable differences, while event-study plots can reveal certain timing problems; neither resolves every time-varying or location-specific alternative. A compelling account states which bias each test probes and which biases remain. Robustness is cumulative knowledge about the vulnerability of a conclusion, not a tally of models that retain statistical significance. Callaway & Sant’Anna (2021) Cinelli & Hazlett (2020)

Selection, Attrition, and Missing Spatial Data

The observed population in a place-based study is produced through recruitment, geocoding, device participation, residence continuity, and outcome follow-up. If people facing housing precarity or severe distress are more likely to relocate or become unreachable, an analysis restricted to participants with two complete observations may underrepresent those most affected by environmental change. Missing GPS observations may concentrate in underground transit, dense buildings, particular devices, or situations in which participants disable tracking. These are potentially informative processes. The investigator should describe sampling at each stage, characterize losses by baseline exposure and outcome predictors, and determine what population remains represented after exclusions. Helbich (2018) Chaix (2009)

Inverse probability of observation weighting and multiple imputation can sometimes reduce bias under assumptions about the missing-data process, availability of sufficient predictors, and model specification. Neither method can recover an entire unobserved displacement process without informative data. When an urban intervention changes who resides in an area, a before–after neighborhood average may change because different people are measured even if no continuing resident's outcome changed. Conversely, resident-cohort analyses may miss benefits or costs borne by newcomers. Both are potentially legitimate estimands, but they require different denominators and follow-up procedures. Hernán & Robins (2016) Ludwig et al. (2012)

Measurement availability can intersect with inequality. Detailed street imagery may favor well-mapped urban corridors, personal monitors may reach people with flexible schedules, and smartphone-based tracking can exclude those unable or unwilling to participate. The representativeness of the exposure measurement process is therefore separable from representativeness of the outcome survey. A study can have a large administrative cohort but geocode rural or informal settlements less accurately. A transparent protocol reports the proportions of unmatched addresses, missing exposure periods, lost participants, and excluded places, together with sensitivity analyses for these patterned omissions. Diez Roux (2001) Kwan (2012)

Spatial Dependence, Statistical Precision, and Multiple Testing

Geographic exposure observations are often correlated. People on neighboring blocks share weather, transportation networks, school districts, and local policies; students within a school share classrooms; residents near the same vacant lot receive a common intervention. Treating each individual's outcome as an independent experimental replicate when assignment occurs at the cluster level can overstate precision. Randomization-based inference or standard errors appropriate to the assignment and spatial error structure should reflect how independent comparisons actually arise. In the Philadelphia greening experiments, the randomized units were clusters of vacant lots rather than hundreds of unrelated individual environmental treatments. South et al. (2018) Branas et al. (2018)

Spatial autocorrelation also affects observational estimation. A pollutant map may have smooth geographic gradients, while distress scores vary with shared regional labor markets and health services. Adding a spatial random effect can improve descriptive fit but may change the target contrast by absorbing variation that the causal exposure also carries. Controlling for latitude, neighborhood fixed effects, or spatial trends should follow a causal account of confounding and the geographic scale of policy, rather than an automatic modeling recipe. Sensitivity across plausible spatial dependence structures strengthens interpretation when the causal estimand remains stable. Diez Roux (2001) Greenland et al. (1999) The spatial causal methods synthesis by Reich and colleagues (2021) explains how structured spatial correlation and exposure spillovers complicate exchangeability and interpretation. Their review underscores the value of designing a spatially explicit intervention model before selecting autocorrelation corrections. Reich et al. (2021)

A second precision problem arises from exposure specification search. Researchers may evaluate several buffer sizes, vegetation products, outcome scales, exposure lags, subsets, and covariate combinations. If only the strongest association is reported, its nominal interval and p-value do not summarize the search process. Prespecifying the main exposure measure and scientifically motivated secondary contrasts, reporting all relevant specifications, and preserving code for geographic derivations address this difficulty. When exploration identifies a promising scale, confirmation in independent locations or periods is more persuasive than relabeling the selected exploratory result as planned. James et al. (2015) Gascon et al. (2015)

Environmental Justice and Distribution of Exposure

Environmental exposure is distributed through political decisions, housing markets, infrastructure investment, racialized exclusion, labor relations, and public service provision. Communities can face higher pollution burdens and fewer accessible amenities while also having fewer resources to relocate. Such patterns are empirical and institutional research objects; individual preferences alone do not explain the geography of unequal opportunity. Residential exposure maps can make spatial inequality visible, but they should distinguish who is exposed, who controls the source, who receives the benefits of intervention, and whose outcomes are measured. A census-area average may conceal important disparities between adjacent blocks and within neighborhoods. Diez Roux (2001) Chaix (2009)

Environmental justice evaluations have at least three questions. Distributive analysis asks which groups bear harms or gain benefits. Procedural analysis asks whose knowledge, priorities, and agency shape decisions. Causal evaluation asks what an actual policy changes for identified populations. These questions intersect without collapsing into a single neighborhood-level score. A greening intervention might reduce fear among existing residents, alter the use of public space, and contribute to economic pressures that change residential composition. Its total distributional impact requires tracking displacement and access alongside symptom change rather than treating an average improvement as a full equity accounting. Branas et al. (2018) South et al. (2018)

Measurement choices themselves distribute recognition. An NDVI-based map values vegetation regardless of usable entrances; park proximity may overstate access for residents who cannot cross dangerous roads; smartphone movement records omit participants with constrained connectivity. Community-derived audits and experience sampling can identify barriers invisible to remote sensing, while calibration against physical instruments helps prevent bias in self-report. Participatory input can refine what counts as an exposure and make unanticipated harms measurable. Its role should be documented as part of construct development and governance, with independent statistical evaluation of intervention effects. Kwan (2018) Hartig et al. (2014)

Causal Generalization and Transportability

A local causal effect is anchored in a population, policy, time, and environment. A vacant-lot remediation trial in Philadelphia may offer evidence for municipalities with similar parcels and governance, but its treatment package may change when translated into a region with scarce water, different land ownership, or a radically different housing market. A housing mobility effect among eligible families in a historical U.S. voucher demonstration may depend on destinations, school systems, and age at move. The external-validity question is therefore substantive: which causal pathways, baseline risks, opportunities, and implementation characteristics would remain similar in a target context? South et al. (2018) Ludwig et al. (2012)

Psychological heterogeneity matters for translation. Children, older adults, people with mobility impairments, and workers whose daily activity extends far from home have different exposure schedules. Greenness measured around a residence may weakly represent a working adult's daylight contacts, while a school-based measure may better represent a child's attention context. Environmental effects can also differ according to baseline distress, perceived safety, climate, public transport, and quality of social infrastructure. Subgroup estimation requires sufficient data and correction for multiplicity, but a theoretically plausible average across groups can still conceal consequential policy differences. Dadvand et al. (2015) Helbich (2018)

Transport should be supported by evidence comparing the source and target settings. Researchers can report the distribution of age, socioeconomic characteristics, baseline exposure, urban form, and intervention fidelity, then assess whether effect modification could plausibly change conclusions. Cross-city studies may provide broader descriptive coverage but remain observational; a collection of locally randomized evaluations can provide stronger identification for multiple settings. No universal mathematical transformation can make a vaguely defined environmental exposure comparable across all social ecologies. The durable unit of cumulative science is the explicitly described intervention–population–outcome contrast. Hernán & Robins (2016) Gascon et al. (2015)

Reproducibility and Open Geographic Science

A reproducible place-based research pipeline begins before estimation. It fixes the geographic reference system, environmental data release, date range, geocoding procedure, raster extraction or network construction, exposure aggregation, outcome coding, and statistical model. The analytic record should retain decisions about buffer distances, missing addresses, map revisions, and historical boundary changes. Without these details, another research team may obtain a different result from the same conceptual variables simply because the underlying geographic processing rules changed. Clear reproducibility documentation is especially important when administrative locations and third-party environmental products are regularly updated. Hystad et al. (2011) James et al. (2015)

Transparency has boundaries when geolocation can reveal personal identity. An open repository can publish simulation code, synthetic example coordinates, variable definitions, summary diagnostics, and restricted-access reproducible procedures without disclosing participants' routes or addresses. Controlled access agreements may permit qualified replication of sensitive data; privacy-preserving spatial perturbation should be quantified because the perturbation itself can affect short-distance exposures. A published result should identify which outputs are reproducible from public material and which depend on controlled data. Openness is a governance design as well as a technical release decision. Kwan (2012) Helbich (2018)

Replication can involve several distinct tasks: reconstructing the original GIS exposure, reproducing the analysis on identical data, estimating the same target effect in another time or place, and comparing outcomes under a newly implemented intervention. Agreement of a single fitted coefficient across two algorithms is weaker than independent evidence under different measurement and selection processes. Field-wide cumulative knowledge improves when null results, sensitivity analyses, departures from preregistration, and detailed intervention descriptions are made available. Inference grows from documented comparability across studies rather than from identical environmental labels. South et al. (2018) Rohrer (2018)

A Scientific Appraisal Framework for Place-Based Claims

A credible environmental causal argument can be evaluated through an ordered set of connected questions. First identify the environmental object and the process through which people encounter it. Next identify the intervention or exposure contrast, target population, psychological outcome, and relevant times. Then examine how the observations represent the exposure and whether error differs by group or outcome. Finally ask what assignment mechanism or identifying assumptions permit counterfactual interpretation. Each answer constrains the next: a perfect statistical adjustment cannot repair an exposure variable that represents the wrong setting, and a highly precise sensor cannot compensate for an implausible comparison group. Hernán & Robins (2016) Kwan (2012)

Source appraisal should distinguish randomized assignment, observational longitudinal contrasts, cross-sectional associations, and narrative or systematic syntheses. Within each class, evidence quality depends on implementation and follow-up, measurement validation, representative sampling, confidence intervals, interference assessment, and compatibility between analytic and substantive units. Sensitivity analysis adds an explicit statement of vulnerability to missing causes. Comparison of published environmental interventions should foreground which policies were changed and what outcome windows were observed rather than awarding a single abstract quality score to all place-based research. South et al. (2018) Cinelli & Hazlett (2020) Gascon et al. (2015)

Scientific synthesis is strongest when it recognizes separable unresolved tasks. Better remote sensing improves physical mapping; mobile records help estimate contact; validated self-report identifies experienced meaning; experiments identify policy effects; longitudinal strategies examine time-dependent exposure; and participatory studies clarify lived constraints. Each supplies one component of the causal claim. Integrated designs can become more persuasive if these components are aligned in a declared research protocol. The result is a program of environmental psychological knowledge that treats places as measurable processes and persons as moving participants in ecological histories. Helbich (2018) Hartig et al. (2014)

Designing a Target Trial of Neighborhood Greening

Consider a city that plans to restore vacant lots in neighborhoods with high levels of neglected land. A target trial would enroll eligible residential clusters before treatment, specify a greening package with maintenance obligations, define a comparison strategy, and identify a primary psychological outcome such as change in validated distress scores over a stated period. The unit of assignment and potential interference zone would be determined by how far visual, social, and behavioral effects are expected to travel. Eligibility might require stable parcel ownership and an agreed implementation schedule; these restrictions would define the population to which the estimate applies. Hernán & Robins (2016) South et al. (2018)

The measurement plan would characterize treated lots with geocoded polygons, documented vegetation and cleanup changes, independent field audits, and the dates when residents could first experience the transformed spaces. Residents' addresses and activity spaces would be observed, with suitable privacy controls, to estimate the distinction between assignment, available exposure, and actual contact. Psychological assessments would occur at specified preintervention and follow-up times. Baseline covariates would include socioeconomic and housing conditions, prior distress, and other exposures with credible confounding roles. The plan would state whether the primary estimand was an assignment effect for baseline residents or an effect on the neighborhood's current population. Kwan (2012) Branas et al. (2018)

Analysis would respect cluster assignment and prespecified contrasts, account for attrition and plausible interference, and report confidence intervals alongside effect estimates and practical significance. Separate mediator analyses could examine perceived safety or outdoor activity, with additional assumptions made explicit. If randomization were infeasible, the design would specify the policy-selection process and a plausible natural-experiment or difference-in-differences comparison before inspecting postintervention results. The general scientific lesson is that the treatment, its geographic implementation, the exposure actually received, and the psychological response must be measured as connected but distinct events. South et al. (2018) Callaway & Sant’Anna (2021) Imai et al. (2010)

A Multimethod Protocol for Cumulative Environmental Exposure

For cumulative exposure research, the observation model should combine dated residential histories, time-varying environmental surfaces, mobility-based contact estimates when feasible, and repeated psychological outcomes. The target may be the consequence of maintaining a specified level of traffic-related pollution during childhood or of receiving a public-space intervention across several seasons. Estimating that target requires deciding whether exposure is measured as an annual mean, a moving average, a peak-weighted history, or a sequence of conditions. The chosen summary must correspond to plausible mechanisms and latency periods; interchangeable exposure aggregations are scientifically implausible when outcomes operate on radically different timescales. Helbich (2018) Hystad et al. (2011)

A validation subsample can collect in-situ measurements or calibrated personal sensors to characterize differences between modeled ambient exposure and actual contact. Repeated EMA can capture changes in affect and perceived safety, while standardized clinical or cognitive instruments can assess longer-term outcomes. Researchers should distinguish the question of accurate contact reconstruction from the question of whether that contact changed functioning. When an exposure influences later daily routines or residential decisions, statistical methods for time-varying treatment and confounding become necessary; simple regression on cumulative average exposure can conceal the process being studied. Robins (1986) Kwan (2018)

Temporal alignment can be tested through planned exposure windows and negative-control periods when appropriate. In a school setting, measurements during class hours may be more relevant to attention than residential nighttime averages; for sleep disturbance, the opposite may be true. Participants who move, stop wearing sensors, or become unreachable require an explicit observation model. Geographic resolution is one part of validity, complemented by behavioral sampling, outcome reliability, causal design, and transparent handling of missing information. Stansfeld et al. (2005) Dadvand et al. (2015) Subiza-Pérez et al. (2025)

Environmental Mixtures and Separable Interventions

Environmental exposures co-occur because infrastructure and urban form produce multiple conditions together. Major roads create noise and traffic emissions; limited shade raises heat exposure; disinvestment can reduce accessible green space while changing perceptions of safety. These conditions may confound one another, act jointly, or belong to the same intervention pathway. Regression models that include every measured environmental variable can yield coefficients that are unstable or that represent implausible changes holding all correlated conditions fixed. A policy-centered research design begins instead with a feasible intervention and examines the combined and potentially separable changes it causes. Werder et al. (2024) Hartig et al. (2014)

Mixture analyses can identify patterns of co-exposure and differential vulnerability but have their own interpretation requirements. A mixture index may summarize disadvantage without representing a manipulable exposure strategy, and statistical feature importance is a predictive quantity unless a suitable causal intervention is defined. Analysts should examine how indicators are correlated, whether there are realistic observations for the proposed independent contrasts, and whether a single source of pollution contributes multiple markers. In the Gulf cohort, observed associations between particulate matter, greenness, and depression demonstrate the importance of jointly characterized environments but do not prove that changing vegetation would cancel an individual particulate exposure effect. Werder et al. (2024) Rohrer (2018)

Factorial or otherwise structured interventions could in principle vary greenery, traffic restrictions, and neighborhood maintenance to test components. Practical constraints may require staged policies or observational mediation designs, each with different assumptions. A valuable finding can also concern the total package: a combined traffic-calming and shade intervention may be more policy-relevant than an unrealistic isolated unit increase in remote-sensed vegetation. The publication should name this choice and resist turning a joint policy effect into several independently established mechanisms. South et al. (2018) Imai et al. (2010)

Contextual Validity Across the Life Course

The environment relevant to a psychological outcome changes with life stage and institutional role. For a young child, home, school, playground, and caregiver travel routes may dominate; for a working adult, commuting, workplace conditions, and household obligations may be central; for a person with reduced mobility, the immediate dwelling and accessible local services may matter most. The same residential buffer can therefore have different validity across groups. Time at risk also differs: an acute emotion measure, a school-year change in attention, and adult outcomes after childhood relocation require separate temporal windows. Dadvand et al. (2015) Chetty et al. (2016)

Developmental research can become more informative when it combines geography with timing of exposure. The MTO administrative follow-up indicates that childhood relocation effects differ by age at move, a finding consistent with sensitivity to duration and developmental period but that does not isolate one environmental component. Longitudinal geographic analyses need residence histories spanning relevant years, records of school changes, and attention to family shocks that induce relocation. A location observed at one wave cannot stand in for an entire developmental ecology without additional evidence. Chetty et al. (2016) Helbich (2018)

Generalization to groups with different mobility or sensory access requires directly testing whether the measurement captures their actual circumstances. Travel-time metrics may depend on walking speed and transit accessibility; measures of restorative space may require accessible paths, perceived welcome, and suitable facilities. If the same geographic index obscures these differences, observed variation in estimated effects can reflect construct mismatch as well as real biological or psychosocial effect modification. A responsible synthesis therefore interprets demographic heterogeneity together with exposure validity and access conditions. Kwan (2018) Diez Roux (2001)

Artificial Sensing and Human–Artificial Research Relations

The Artificial Era introduces autonomous or semi-autonomous systems into environmental observation: machine-learning models classify satellite images, portable devices estimate movement and sound, and computational tools reconstruct exposure from heterogeneous geographic records. Their scientific contribution lies in identifiable operations—measurement, prediction, classification, calibration, and quality control. An artificial system can help estimate vegetation at a location or detect a change in traffic noise, but the resulting numerical output inherits training-data coverage, sensor error, domain shift, and assumptions encoded in the feature definitions. Human psychological claims additionally require outcome validity, causal identification, and an account of lived contact. Hystad et al. (2011) Helbich (2018)

Human–artificial relations may themselves change activity spaces. Navigation systems influence routes, remote work alters commuting exposure, recommendation interfaces shape which parks or venues people discover, and conversational tools can affect experiences of social support or perceived information access. These are plausible research questions rather than established effects of a single artificial technology. Their study should specify the intervention, the human behavioral outcome, and the environmental exposures created or avoided by changed routines. Artificial agency is relevant as a component of the social and informational ecology when it changes the pathways through which people encounter physical environments. Kwan (2012) Hernán & Robins (2016)

A defensible cross-domain research program can compare computer-predicted exposure against validated sensors, assess subgroup errors in mapping and recommendation systems, and evaluate whether algorithmically guided travel changes contact with heat, pollution, noise, or safe public spaces. Such evaluation keeps measurement performance distinct from downstream psychological effects and avoids assuming that an apparently intelligent system has observed the lived meaning of place. Claims about artificial psychological organization belong to a separate theoretical and empirical research question. In this monograph, artificial systems are considered as instruments and potential determinants of environmental contact, within the broader historical transformation From Homo to Artificial. Rohrer (2018) Kwan (2018)

Unresolved Research Problems and the Next Empirical Agenda

The first unresolved problem is identifying the correct geography and time scale for a causal process whose boundaries vary across persons and circumstances. Dense mobile sensing can reduce uncertainty about location while introducing differential participation and threats to privacy. The second is construct equivalence: remote-sensed vegetation, personally experienced restoration, and usable public space respond to different interventions. A third is selection over the life course, where prior psychological functioning influences where a person moves and what environments are later encountered. Future studies need validation data, repeated measures, and intervention protocols that jointly describe these processes. Kwan (2012) Helbich (2018) Robins (1986) Christensen and colleagues' 2025 review of 33 geographically informed momentary-assessment studies shows that intensive geographic observation has developed more quickly than designs capable of identifying temporal environmental effects; this imbalance should guide future study design. Christensen et al. (2025)

A fourth problem is the transport of causal effects beyond the original neighborhoods, cities, and countries. Place-based policies are implemented within governance, housing, climate, and service systems that change treatment versions and exposure opportunities. A fifth is evaluation of spillovers and displacement: improving one parcel may influence adjacent blocks, visitor flows, housing costs, and population composition. A sixth is the differentiation of mixed interventions, such as greening that also cleans and maintains an area. These problems require linked geographic, psychosocial, institutional, and longitudinal evidence rather than a universal environmental coefficient. South et al. (2018) Ludwig et al. (2012) Werder et al. (2024)

The next empirical agenda should include repeated measurement validation, transparent sensitivity analysis, registered natural-experiment and randomized designs, multiple outcome horizons, baseline resident tracking, and explicit distributional assessment. Long-term research infrastructure should preserve boundary versions and maps in privacy-respecting form so findings can be reanalyzed as measurement methods develop. The strongest future synthesis will compare studies with genuinely comparable interventions, exposure measures, and populations, while explaining why otherwise similar environmental labels produced different findings. Such a program is capable of accumulating trustworthy knowledge about environmental conditions and psychological functioning without treating observational proximity as self-evident causation. Gascon et al. (2015) Cinelli & Hazlett (2020)

Conclusion

Environmental exposure measurement and place-based causal inference form a single scientific problem because a causal effect depends upon the environmental condition that the research design has successfully defined and measured. A recorded home, census boundary, satellite index, ambient monitor, subjective rating, or daily trajectory each constitutes evidence about a particular aspect of context. Their strengths and weaknesses determine what psychological processes can plausibly be studied. Inference begins with the research object: a defined experience or policy change, a population, a period of contact, an outcome, and alternative conditions under which that outcome would be compared. Diez Roux (2001) Kwan (2012) Hernán & Robins (2016)

The empirical record establishes several instructive contrasts. Cluster-randomized restoration of vacant land supports conclusions about implemented environmental intervention packages, while randomized voucher offers establish effects of a housing policy and reveal the importance of childhood exposure timing. Longitudinal school greenness and relocation studies offer more detailed temporal information than one-time maps, yet must account for spatial selection. Cross-sectional evidence on pollution and greenery illustrates the importance of environmental mixtures without isolating a causal intervention. These findings become cumulative when their units of assignment, observed exposures, outcome measures, time scales, and identifying assumptions are kept explicit. South et al. (2018) Ludwig et al. (2012) Chetty et al. (2016) Dadvand et al. (2015) Werder et al. (2024) Subiza-Pérez et al. (2025)

The durable methodological result is a coordinated research architecture: mechanism-informed exposure definitions, spatial and temporal validation, explicit causal contrasts, credible assignment or conditional identification, sensitivity to bias, assessment of interference and generalization, and transparent reproducible documentation. Environmental justice extends this architecture by identifying whose environments are measured, whose outcomes improve, and who can access or remain in changed places. Artificial measurement systems add new observational capacity and new validation responsibilities. The field advances when increasingly precise descriptions of environments are matched by equally precise accounts of what changing those environments does to psychological life. Branas et al. (2018) Rohrer (2018) Helbich (2018)

References

  • Bardhan, M., Li, F., Browning, M. H. E. M., Dong, J., Zhang, K., Yuan, S., İnan, H. E., McAnirlin, O., Dagan, D. T., Maynard, A., Thurson, K., Zhang, F., Wang, R., & Helbich, M. (2024). From space to street: A systematic review of the associations between visible greenery and bluespace in street view imagery and mental health. Environmental Research, 263(Part 3), 120213. https://doi.org/10.1016/j.envres.2024.120213
  • Branas, C. C., South, E., Kondo, M. C., Hohl, B. C., Bourgois, P., Wiebe, D. J., & MacDonald, J. M. (2018). Citywide cluster randomized trial to restore blighted vacant land and its effects on violence, crime, and fear. Proceedings of the National Academy of Sciences, 115(12), 2946–2951. https://doi.org/10.1073/pnas.1718503115
  • Callaway, B., & Sant’Anna, P. H. C. (2021). Difference-in-differences with multiple time periods. Journal of Econometrics, 225(2), 200–230. https://doi.org/10.1016/j.jeconom.2020.12.001
  • Chaix, B. (2009). Geographic life environments and coronary heart disease: A literature review, theoretical contributions, methodological updates, and a research agenda. Annual Review of Public Health, 30, 81–105. https://doi.org/10.1146/annurev.publhealth.031308.100158
  • Chetty, R., Hendren, N., & Katz, L. F. (2016). The effects of exposure to better neighborhoods on children: New evidence from the Moving to Opportunity experiment. American Economic Review, 106(4), 855–902. https://doi.org/10.1257/aer.20150572
  • Christensen, R., Constable Fernandez, C., Topalian, N., Vaughan, L., Krenz, K., Pitman, A., & Fett, A.-K. (2025). A systematic review of geographically informed ecological momentary assessment studies on the place-based correlates of mental health, substance use and wellbeing. Landscape and Urban Planning, 264, 105487. https://doi.org/10.1016/j.landurbplan.2025.105487
  • Cinelli, C., & Hazlett, C. (2020). Making sense of sensitivity: Extending omitted variable bias. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 82(1), 39–67. https://doi.org/10.1111/rssb.12348
  • Dadvand, P., Nieuwenhuijsen, M. J., Esnaola, M., Forns, J., Basagaña, X., Alvarez-Pedrerol, M., Rivas, I., López-Vicente, M., De Castro Pascual, M., Su, J., Jerrett, M., Querol, X., & Sunyer, J. (2015). Green spaces and cognitive development in primary schoolchildren. Proceedings of the National Academy of Sciences, 112(26), 7937–7942. https://doi.org/10.1073/pnas.1503402112
  • Diez Roux, A. V. (2001). Investigating neighborhood and area effects on health. American Journal of Public Health, 91(11), 1783–1789. https://doi.org/10.2105/ajph.91.11.1783
  • Gascon, M., Triguero-Mas, M., Martínez, D., Dadvand, P., Forns, J., Plasència, A., & Nieuwenhuijsen, M. J. (2015). Mental health benefits of long-term exposure to residential green and blue spaces: A systematic review. International Journal of Environmental Research and Public Health, 12(4), 4354–4379. https://doi.org/10.3390/ijerph120404354
  • Greenland, S., Pearl, J., & Robins, J. M. (1999). Causal diagrams for epidemiologic research. Epidemiology, 10(1), 37–48. https://pubmed.ncbi.nlm.nih.gov/9888278/
  • Hartig, T., Mitchell, R., de Vries, S., & Frumkin, H. (2014). Nature and health. Annual Review of Public Health, 35, 207–228. https://doi.org/10.1146/annurev-publhealth-032013-182443
  • Helbich, M. (2018). Toward dynamic urban environmental exposure assessments in mental health research. Environmental Research, 161, 129–135. https://doi.org/10.1016/j.envres.2017.11.006
  • Hernán, M. A., & Robins, J. M. (2016). Using big data to emulate a target trial when a randomized trial is not available. American Journal of Epidemiology, 183(8), 758–764. https://doi.org/10.1093/aje/kwv254
  • Hystad, P., Setton, E., Cervantes, A., Poplawski, K., Deschenes, S., Brauer, M., van Donkelaar, A., Lamsal, L., Martin, R., Jerrett, M., & Demers, P. (2011). Creating national air pollution models for population exposure assessment in Canada. Environmental Health Perspectives, 119(8), 1123–1129. https://doi.org/10.1289/ehp.1002976
  • Imai, K., Keele, L., & Tingley, D. (2010). A general approach to causal mediation analysis. Psychological Methods, 15(4), 309–334. https://doi.org/10.1037/a0020761
  • James, P., Banay, R. F., Hart, J. E., & Laden, F. (2015). A review of the health benefits of greenness. Current Epidemiology Reports, 2(2), 131–142. https://doi.org/10.1007/s40471-015-0043-7
  • Kondo, M. C., Fluehr, J. M., McKeon, T., & Branas, C. C. (2018). Urban green space and its impact on human health. International Journal of Environmental Research and Public Health, 15(3), 445. https://doi.org/10.3390/ijerph15030445
  • Kwan, M.-P. (2012). The uncertain geographic context problem. Annals of the Association of American Geographers, 102(5), 958–968. https://doi.org/10.1080/00045608.2012.687349
  • Kwan, M.-P. (2018). The neighborhood effect averaging problem (NEAP): An elusive confounder of the neighborhood effect. International Journal of Environmental Research and Public Health, 15(9), 1841. https://doi.org/10.3390/ijerph15091841
  • Ludwig, J., Duncan, G. J., Gennetian, L. A., Katz, L. F., Kessler, R. C., Kling, J. R., & Sanbonmatsu, L. (2012). Neighborhood effects on the long-term well-being of low-income adults. Science, 337(6101), 1505–1510. https://doi.org/10.1126/science.1224648
  • Pearl, J. (1995). Causal diagrams for empirical research. Biometrika, 82(4), 669–688. https://doi.org/10.1093/biomet/82.4.669
  • Reich, B. J., Yang, S., Guan, Y., Giffin, A. B., Miller, M. J., & Rappold, A. (2021). A review of spatial causal inference methods for environmental and epidemiological applications. International Statistical Review, 89(3), 605–634. https://doi.org/10.1111/insr.12452
  • Robins, J. M. (1986). A new approach to causal inference in mortality studies with a sustained exposure period—Application to control of the healthy worker survivor effect. Mathematical Modelling, 7(9–12), 1393–1512. https://doi.org/10.1016/0270-0255(86)90088-6
  • Rohrer, J. M. (2018). Thinking clearly about correlations and causation: Graphical causal models for observational data. Advances in Methods and Practices in Psychological Science, 1(1), 27–42. https://doi.org/10.1177/2515245917745629
  • Rosenbaum, P. R., & Rubin, D. B. (1983). The central role of the propensity score in observational studies for causal effects. Biometrika, 70(1), 41–55. https://doi.org/10.1093/biomet/70.1.41
  • Song, I., & Luan, H. (2026). Ecological context modifies estimated effects of residential greenspace on mental illness mortality: A case study of Washington State, USA. International Journal of Health Geographics. Advance online publication. https://doi.org/10.1186/s12942-026-00497-z
  • South, E. C., Hohl, B. C., Kondo, M. C., MacDonald, J. M., & Branas, C. C. (2018). Effect of greening vacant land on mental health of community-dwelling adults: A cluster randomized trial. JAMA Network Open, 1(3), e180298. https://doi.org/10.1001/jamanetworkopen.2018.0298
  • Stansfeld, S. A., Berglund, B., Clark, C., Lopez-Barrio, I., Fischer, P., Ohrström, E., Haines, M. M., Head, J., Hygge, S., van Kamp, I., & Berry, B. F. (2005). Aircraft and road traffic noise and children's cognition and health: A cross-national study. The Lancet, 365(9475), 1942–1949. https://doi.org/10.1016/S0140-6736(05)66660-3
  • Subiza-Pérez, M., Mebrahtu, T. F., Krenz, K., Watmuff, A., Yang, T., Vaughan, L., Wright, J., & McEachan, R. R. C. (2025). A good move for health? Analyzing urban exposure trajectories of residential relocation and mental health in populations in Bradford. Environmental Epidemiology, 9(3), e397. https://doi.org/10.1097/EE9.0000000000000397
  • Sun, L., & Abraham, S. (2021). Estimating dynamic treatment effects in event studies with heterogeneous treatment effects. Journal of Econometrics, 225(2), 175–199. https://doi.org/10.1016/j.jeconom.2020.09.006
  • VanderWeele, T. J., & Ding, P. (2017). Sensitivity analysis in observational research: Introducing the E-value. Annals of Internal Medicine, 167(4), 268–274. https://doi.org/10.7326/M16-2607
  • Werder, E., Lawrence, K., Deng, X., Jackson, W. B., II, Christenbury, K., Buller, I., Engel, L., & Sandler, D. (2024). Residential air pollution, greenspace, and adverse mental health outcomes in the U.S. Gulf Long-term Follow-up Study. Science of the Total Environment, 946, 174434. https://doi.org/10.1016/j.scitotenv.2024.174434