Treatment Geometry and Causal Identification with Earth Observation Data

Anna Josephson; Elinor Benami; Ella Kirchner; Gina Maskell; Jeffrey D. Michler; Kunwar Singh; Michael J. Cecil; Patrick Behrer; Robert Heilmayr; Sydney Gourlay

REVIEW 2 major objections 7 minor 7 references

The treated area you draw on a map decides what causal effect you estimate.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Defines treatment geometry and provides a protocol for making spatial and temporal exposure choices explicit in geospatial impact evaluations.

T0 review reviewed 2026-08-01 challenge →

load-bearing objection A solid synthesis chapter that gives geospatial impact evaluation a useful vocabulary and a sensible default protocol, but it's a framework, not a new result; a referee can help clean it up and temper its 'credibility' claims. the 2 major comments →

arxiv 2607.19908 v1 pith:RKVMIAP5 submitted 2026-07-22 econ.EM

Treatment Geometry and Causal Identification with Earth Observation Data

Jeffrey D. Michler , Anna Josephson , Elinor Benami , Patrick Behrer , Michael J. Cecil , Sydney Gourlay , Robert Heilmayr , Ella Kirchner

show 2 more authors

Gina Maskell Kunwar Singh

This is my paper

classification econ.EM

keywords treatment geometrygeospatial impact evaluationcausal identificationearth observation dataexposure assignmentspilloversmeasurement errorresearch design

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This chapter introduces treatment geometry—the spatial and temporal footprint through which exposure to a treatment is represented in data—as a core analytical concept for geospatial impact evaluation. It argues that choices about polygons, buffers, time windows, and assignment rules are not technical details: they define the estimand, the comparison group, and the main threats to causal identification. The paper decomposes treatment geometry into four elements—source, treated unit, pathway, and temporal window—and provides a protocol for making these choices explicit, justified, and reproducible. A sympathetic reader would take away that the credibility of causal estimates from satellite-linked studies depends less on data volume than on how exposure is encoded.

Core claim

The chapter's central claim is that any geospatial impact evaluation implicitly chooses a treatment geometry—the spatial and temporal footprint through which exposure is encoded—and that this choice is itself part of the identification strategy. Because multiple reasonable geometries can exist for the same research question, the paper proposes decomposing the geometry into the source of treatment, the treated unit, the pathway of exposure, and the temporal window. These elements jointly determine which units are treated, which serve as controls, what variation identifies the effect, and what assumptions are required for causal interpretation. Rather than prescribing a single correct geometry

What carries the argument

Treatment geometry is the central object: the spatial and temporal footprint through which exposure to a policy, intervention, hazard, or environmental condition is represented. It is operationalized by decomposing it into four elements—source, treated unit, pathway, and temporal window—which together define a defensible mapping from the underlying causal process to an empirical treatment variable. The concept carries the argument by forcing transparency about assumptions that are often left implicit, such as where exposure becomes zero, how exposure travels, and which temporal window is relevant.

Load-bearing premise

The framework assumes researchers can specify the causal pathway—source, treated unit, pathway, and temporal window—with enough confidence to choose among alternative geometries.

What would settle it

A meta-analysis of published geospatial impact evaluations that report multiple treatment geometries would falsify the core claim if it showed that estimates are rarely sensitive to changes in buffers, time windows, or aggregation units in typical settings.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

Researchers should define and document treatment geometry before estimation, treating buffer distances, aggregation rules, and time windows as modeling assumptions rather than defaults.
Sensitivity analysis across alternative geometries becomes a core diagnostic: if results depend on small changes in buffers, thresholds, or windows, the causal claim is correspondingly fragile.
Spillovers should be built into comparison-group construction—by excluding, modeling, or explicitly estimating indirectly treated units—rather than treated only as a post-estimation robustness check.
Spatial anonymization of survey coordinates limits which research questions are answerable; high-resolution exposure measures are more sensitive to coordinate displacement than smooth, coarse weather products.
The same underlying data can support different estimands: a binary treatment answers a different question than a continuous dose-response measure, and researchers should align the geometry with the intended causal object.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

A natural extension is to formalize treatment geometry as a reporting standard for geospatial impact evaluations, similar to pre-analysis plans, so that readers can see which geometry choices were made and why.
The framework could be applied to settings beyond satellite-linked socioeconomic studies, including environmental epidemiology and conservation science, where exposure surfaces are similarly constructed from buffers, plumes, or travel-time catchments.
A testable extension would be a benchmark exercise where several plausible treatment geometries are applied to the same research question to quantify how much estimates and inference vary across reasonable choices.
The four-element decomposition may also serve as a diagnostic tool for machine-learning-based exposure models, indicating where learned exposure surfaces conflict with the assumed source, pathway, or temporal window.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Desk Editor's Note

A solid synthesis chapter that gives geospatial impact evaluation a useful vocabulary and a sensible default protocol, but it's a framework, not a new result; a referee can help clean it up and temper its 'credibility' claims.

read the letter

Colleague,

Here's the short version: it's a competent, useful synthesis chapter, not a research contribution. It gives the field a useful name—treatment geometry—and a four-part decomposition (source, treated unit, pathway, temporal window) that genuinely clarifies how exposure definitions shape identification. If I were teaching geospatial impact evaluation, I'd assign it. The six case studies are well chosen and illustrate the main tradeoffs: spillovers, directional transport, sparse data, timing, intensive versus extensive margins, and georegistration. The chapter is honest about what it builds on—MAUP, uncertain geographic context, SUTVA/interference, nonclassical measurement error. No inflated claims there.

What it doesn't do is produce new theory or new empirical evidence. The central claim—geometry choices matter for identification—is asserted and illustrated, not proven. For a framework chapter, that's fine; the burden is coherence and usefulness, and it meets those. The protocol in Box 1 is a sensible checklist; it won't tell you which geometry is right. The stress-test worry about underdetermination is real but not fatal: if a researcher's causal prior is wrong, sensitivity analysis over alternative geometries can be stable and still biased. The paper itself says the exposure process is 'only partially observed' and 'rarely a single correct geometry,' so it doesn't claim to solve that; it claims to make assumptions visible, testable, reproducible. That's a meaningful, modest contribution. The one thing a referee should push on is the line in the abstract and conclusion about 'strengthening credibility'—that overstates what transparency can deliver. Suggest rewording to 'reduces the risk of arbitrary choices' or 'improves interpretability.'

On the mechanics: the manuscript is rougher than it should be. There are 'Figure XX' / 'Figure YY' / 'Box XX' placeholders, typos ('Operationationalizing,' 'anAssociate,' 'Vincente-Serrano,' 'Lui'), a duplicated Heilmayr bio, and a citation mismatch (Chatterjee 2020 in text vs Chatterjee 2022 in references; Petruccelli et al. 2026 and Josephson et al. 2026 are listed as forthcoming, which is fine if confirmed). None of this is substantive, but it needs a careful edit.

I'd send this to peer review. The subfield will use it as a reference, and a referee with applied spatial experience can help scope the claims and clean up the presentation. I'd accept after moderate revision rather than desk-reject.

Referee Report

2 major / 7 minor

Summary. This manuscript introduces the concept of "treatment geometry" for geospatial impact evaluation (GIE): the spatial and temporal footprint through which exposure to a treatment is represented in data. The chapter decomposes treatment geometry into four components—source, treated unit, pathway, and temporal window—and discusses how choices about these components interact with data resolution, aggregation, temporal alignment, treatment intensity, boundary uncertainty, and directional transport. It connects geometry choices to identification concepts such as estimand clarity, comparison-group construction, SUTVA, misclassification, and nonclassical measurement error. The second half offers six case studies from conservation, air pollution, agricultural shocks, conflict, and survey georegistration, and closes with a protocol (Box 1) for documenting and stress-testing geometry choices before estimation. The contribution is a conceptual synthesis and practical checklist rather than a formal theorem or an empirical validation of the proposed framework.

Significance. If accepted at face value, the framework would give applied researchers a useful vocabulary for decisions that are often treated as technical preprocessing details—buffer distances, pixel-extraction rules, temporal windows, coordinate displacement—and would connect those decisions to identification. The manuscript draws on a broad and current literature, and the protocol is a valuable transparency device. Its stated goal is not to prescribe one correct geometry but to make geometry choices explicit, defensible, and reproducible. That is a real contribution. However, the paper's central value claim—that explicit treatment geometry can "strengthen the credibility of causal inference"—is asserted and illustrated rather than demonstrated. The case studies are retrospective illustrations, not tests of the protocol, and the framework provides no independent way to adjudicate between a correct and a uniformly misspecified set of geometry assumptions. With a recalibrated claim and one worked example that includes external validation or falsification, the chapter would be a solid methods contribution.

major comments (2)

[Abstract; Section 6, Box 1 (Protocol)] The core claim that explicitly defining treatment geometry "can strengthen the credibility of causal inference" is stronger than the protocol supports. Box 1 asks researchers to document source, unit, pathway, temporal window, and to report sensitivity to alternatives (Q10), but it supplies no independent criterion for adjudicating among geometries when the true exposure process is only partially observed, as Section 6 concedes. Sensitivity over a set of misspecified geometries can be stable while all estimates remain biased; for example, in Case Study 2, if the wind fields used to build the pollution geometry are wrong, varying cone angles will not reveal the error. The manuscript should either (a) add a validation or falsification step—such as comparing the preferred geometry to ground-truth monitors, administrative event records, or a pre-registered hold-out prediction target—or (b) r
[Section 5, Table 1] The six case studies are retrospective narratives assembled after the fact; none is a prospective application of Box 1, and none shows that following the protocol would have corrected or revealed geometry-induced bias. Since the manuscript's value rests on the protocol being actionable, a worked example that implements the full protocol—with pre-specified alternatives, a decision rule, and a falsification exercise—would materially support the central claim. At minimum, the text should state explicitly that the cases are illustrative of the conceptual taxonomy and are not evidence for the protocol's effectiveness.

minor comments (7)

[Section 2 heading] "Operationationalizing" is a typo for "Operationalizing."
[Section 3.2.1 and Section 6] There are two boxes both numbered Box 1: "Collecting Ground Reference Data" in Section 3.2.1 and "A Protocol for Defining Treatment Geometry" at the end of Section 6. These should be renumbered and cross-referenced consistently.
[Section 3.1, Figure XX; Section 3.2.2, Figure YY; Further Resources, Table XX] Several placeholders remain in the text: Figure XX, Figure YY, and Table XX. These need to be resolved before publication.
[Case Study 4, Key risks] "aligning exposure to data availability rather than to casual process" should read "causal process."
[Case Study 5 and References] The in-text citation "Lui et al., 2022" does not match the reference "Liu, Y., et al. (2022)" in the bibliography. Also, "Vincente-Serrano" in the text should be "Vicente-Serrano" to match the reference list.
[Contributors' Biographies] Robert Heilmayr's biography appears twice, with slightly different affiliations. This duplication should be removed.
[References] Some entries are working papers or preprints (e.g., Grosset-Touba et al. 2024; Jordán and Heilmayr 2024; Pignède 2025). If the chapter is intended as a permanent reference, the authors should note whether these have since been peer reviewed or update the citations accordingly.

Circularity Check

0 steps flagged

No significant circularity; the chapter is a synthesis/protocol, not a derivation whose outputs are built into its inputs.

full rationale

The paper is a methods-framework chapter rather than an empirical derivation. It introduces 'treatment geometry' as a term for the spatiotemporal footprint of treatment and decomposes it into source, unit, pathway, and temporal window. The claimed value—that making geometry explicit strengthens the credibility of causal inference—is argued through synthesis of an external literature (Kwan, 2012; Goodchild, 1992; Gotway and Young, 2002; Keele and Titiunik, 2015, among many others) and through illustrative case studies. No parameter is fitted and then renamed as a prediction; no formal result is derived whose conclusion equals its assumptions; no self-cited uniqueness theorem is invoked to forbid alternatives. The self-citations that appear (Michler et al., 2022; Josephson et al., 2026; Josephson and Michler, 2024) are used as supporting examples about GPS displacement, SDL, and weather-data calibration; they are not load-bearing in the sense that the framework collapses without them. Relevant limitation passages—'In many geospatial applications, the true exposure process is only partially observed' and 'Robustness to arbitrary alternatives is not sufficient to prove identification'—show that the authors themselves disclaim the stronger claim that sensitivity analysis can certify a correct geometry. The skeptic's underdetermination concern is an evidential limitation, not a circularity: the framework's central claim is not proven equivalent to its inputs by construction. Accordingly, no circular steps are identified and the score is 0.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The paper contributes a framework, not a derivation. No free parameters or new entities are introduced. The load-bearing premises are domain assumptions about representability and mechanism knowledge.

axioms (2)

domain assumption The true exposure process can be represented by a finite set of spatial and temporal objects (points, polygons, lines, rasters, time windows).
The entire framework depends on discretizing continuous processes into geometry objects; if exposure is fundamentally continuous and cannot be captured by such objects, the protocol is inapplicable. See Section 2.2.1.
domain assumption Researchers have sufficient knowledge of the causal mechanism to choose and justify a treatment geometry.
Section 2.1 requires specifying source, unit, pathway, and temporal window; without mechanism knowledge, any geometry is arbitrary and the protocol's guidance cannot validate it.

reviewed 2026-08-01 · how reviews work

0 comments

Cite this review

Pith. "Pith review of Treatment Geometry and Causal Identification with Earth Observation Data." pith.science (2026). https://pith.science/paper/RKVMIAP5

@misc{pith2026260719908,
  author       = {Pith},
  title        = {Pith review of: Treatment Geometry and Causal Identification with Earth Observation Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RKVMIAP5}},
  note         = {Machine review of arXiv:2607.19908}
}

Share X Bluesky LinkedIn Reddit HN

read the original abstract

A central task in conducting impact evaluations is determining who or what was exposed to a treatment, when, and to what degree. These questions can be especially complex in geospatial settings, where many reasonable definitions of exposure may exist. This chapter introduces treatment geometry as a core concept in geospatial impact evaluation (GIE): the spatial and temporal footprint of a treatment as represented in data. How this footprint is defined shapes identification strategies and the credibility of causal inference. Drawing on cases spanning the air pollution, wildfire, forest policy, infrastructure, pest, and food security literature, the chapter provides practical guidance on navigating key tradeoffs (including spatial resolution, temporal alignment, spillovers, and boundary uncertainty) that arise when translating real-world interventions into analyzable data. Rather than prescribing a single best approach, the chapter equips researchers with a framework for diagnosing which geometry decisions may matter most in their context, closing with synthesis questions to help readers navigate these decisions.

discussion (0)

Reference graph

Works this paper leans on

7 extracted references · 1 linked inside Pith

[1]

Ahvo, A., Heino, M., Sandström, V., Chrisendo, D., Jalava, M., & Kummu, M. (2023). Agricultural input shocks affect crop yields more in the high-yielding areas of the world. Nature Food 4: 1037–1046. Alix-Garcia, J., & Millimet, D. L. (2023). Remotely incorrect? Accounting for nonclassical measurement error in satellite data on deforestation. Journal of t...

2023
[408]

Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., & Moore, R. (2017). Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment 202: 18-27. Gotway, C. A., & Young, L. J. (2002). Combining incompatible spatial data. Journal of the American Statistical Association 97(458): 632-648. Grosset-Touba, F....

2017
[1034]

Crop calendar

Food and Agriculture Organization (FAO). Crop calendar. FAO. Foody, G. M. (2002). Status of land cover classification accuracy assessment. Remote Sensing of Environment 80(1): 185-201. Fritz, S., See, L., Perger, C., McCallum, I., Schill, C., Schepaschenko, D., Duerauer, M., Karner, M., Dresel, C., Laso-Bayas, J-C., Lesiv, M., Moorthy, I., Salk, C.F., Dan...

2002
[2013]

Political Analysis 21: 42–47

Fisheries Management. Political Analysis 21: 42–47. Li, J., & Heap, A. D. (2014). Spatial interpolation methods applied in the environmental sciences: A review. Environmental Modelling & Software 53: 173-189. Liepa, A., Thiel, M., Taubenböck, H., Klein, D., Steffan-Dewenter, I., Peters, M. K., Schönbrodt- Stitt, S., Otte, I., Landmann, T., Khan, Z. R., Ob...

Pith/arXiv arXiv 2014
[2025]

Access restricted/requires registration World Bank LSMS Mult-topic surveys on poverty and living conditions Selected household coordinates Varies by location 43 countries Access generally restricted to geospatial data. Data and documentation available on the World Bank Microdata Library: https://microdata.worldbank.org World Bank Living Standard Measureme...

2008
[4749]

Baragwanath, K., & Shinde, S. (2025). Beyond the canopy: How satellite data detection thresholds influence policy evaluation and deforestation behavior. Journal of Environmental Economics and Management 134: 103219. Blackman, A., Leguízamo, E., & Villalobos, L. (2024). Points, cells, or polygons? On the choice of spatial units in forest conservation polic...

2025
[5476]

von Uexkull, N., Croicu, M., Fjelde, H., & Buhaug, H. (2016). Civil conflict sensitivity to growing- season drought. Proceedings of the National Academy of Sciences 113(44): 12391– 12396. Warren, J. L., Perez-Heydrich, C., Burgert, C. R., & Emch, M. E. (2015). Influence of Demographic and Health Survey point displacements on distance-based analyses. Spati...

2016

This paper was first reviewed by deepseek-v4-flash on August 1, 2026.

[1] [1]

Ahvo, A., Heino, M., Sandström, V., Chrisendo, D., Jalava, M., & Kummu, M. (2023). Agricultural input shocks affect crop yields more in the high-yielding areas of the world. Nature Food 4: 1037–1046. Alix-Garcia, J., & Millimet, D. L. (2023). Remotely incorrect? Accounting for nonclassical measurement error in satellite data on deforestation. Journal of t...

2023

[2] [408]

Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., & Moore, R. (2017). Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment 202: 18-27. Gotway, C. A., & Young, L. J. (2002). Combining incompatible spatial data. Journal of the American Statistical Association 97(458): 632-648. Grosset-Touba, F....

2017

[3] [1034]

Crop calendar

Food and Agriculture Organization (FAO). Crop calendar. FAO. Foody, G. M. (2002). Status of land cover classification accuracy assessment. Remote Sensing of Environment 80(1): 185-201. Fritz, S., See, L., Perger, C., McCallum, I., Schill, C., Schepaschenko, D., Duerauer, M., Karner, M., Dresel, C., Laso-Bayas, J-C., Lesiv, M., Moorthy, I., Salk, C.F., Dan...

2002

[4] [2013]

Political Analysis 21: 42–47

Fisheries Management. Political Analysis 21: 42–47. Li, J., & Heap, A. D. (2014). Spatial interpolation methods applied in the environmental sciences: A review. Environmental Modelling & Software 53: 173-189. Liepa, A., Thiel, M., Taubenböck, H., Klein, D., Steffan-Dewenter, I., Peters, M. K., Schönbrodt- Stitt, S., Otte, I., Landmann, T., Khan, Z. R., Ob...

Pith/arXiv arXiv 2014

[5] [2025]

Access restricted/requires registration World Bank LSMS Mult-topic surveys on poverty and living conditions Selected household coordinates Varies by location 43 countries Access generally restricted to geospatial data. Data and documentation available on the World Bank Microdata Library: https://microdata.worldbank.org World Bank Living Standard Measureme...

2008

[6] [4749]

Baragwanath, K., & Shinde, S. (2025). Beyond the canopy: How satellite data detection thresholds influence policy evaluation and deforestation behavior. Journal of Environmental Economics and Management 134: 103219. Blackman, A., Leguízamo, E., & Villalobos, L. (2024). Points, cells, or polygons? On the choice of spatial units in forest conservation polic...

2025

[7] [5476]

von Uexkull, N., Croicu, M., Fjelde, H., & Buhaug, H. (2016). Civil conflict sensitivity to growing- season drought. Proceedings of the National Academy of Sciences 113(44): 12391– 12396. Warren, J. L., Perez-Heydrich, C., Burgert, C. R., & Emch, M. E. (2015). Influence of Demographic and Health Survey point displacements on distance-based analyses. Spati...

2016

Treatment Geometry and Causal Identification with Earth Observation Data

The reading

Core claim §

What carries the argument

Load-bearing premise §

What would settle it §

If this is right

Where Pith is reading between the lines

discussion (0)

Core claim

Load-bearing premise

What would settle it