Research Design (aaag-research-design)
The Annals spans four areas and accepts many methodologies, but is demanding about each. The design
must credibly connect the geographic argument (aaag-theory-building) to the evidence, and must take
space and scale seriously — spatial dependence, the MAUP, projection, and sampling are design
issues, not afterthoughts. This skill is mode-aware: pick the section that matches your work.
When to trigger
- Specifying identification, sampling, case selection, or measurement
- A reviewer questioned spatial autocorrelation, scale/MAUP, edge effects, validation, or a confound
- Justifying why the design adjudicates the rival account from
aaag-literature-positioning
Spatial / quantitative analysis & GIScience
- Take space seriously. Test and model spatial dependence (Moran's I, spatial lag/error,
GWR/MGWR where heterogeneity is the point); state how the MAUP / scale could change conclusions.
- Geography of the data. Document projection/CRS, areal units, edge effects, and the support of
measurements; spatial sampling and its biases.
- Inference. Cluster or use spatial SEs at the right level; for spatial autocorrelation, report
diagnostics; for prediction, use spatially-aware cross-validation (blocked/spatial CV), not random folds.
Remote sensing / physical-environmental
- Measurement validity. Sensor/resolution choices, atmospheric/geometric correction, and ground
truth; quantify accuracy (confusion matrix, kappa/F1, RMSE) with an independent validation sample.
- Process linkage. Tie observed pattern to an earth-surface process and its scale; state the
uncertainty budget end to end.
Qualitative / human-geography
- Case selection by design logic (typical, extreme, paired, regional contrast) — say what the case
is a case of. Convenience is not a rationale.
- Positionality, reflexivity, and rigor appropriate to the method (ethnography, interviews,
archives, discourse/textual analysis); state how interpretations were checked.
- Source/field transparency: plan how fieldnotes, interviews, and archives are documented and cited
(see
aaag-transparency-and-data), including consent and geoprivacy.
Nature-society / mixed methods
- Integrate, don't staple. Specify how the biophysical and social strands inform one another
(e.g., land-change observation + livelihood interviews), and how convergence/divergence is handled.
The adjudication test (Annals-specific)
For the single strongest rival explanation, write: "If the rival held rather than my argument, the
[spatial pattern / measurements / accounts] would look like ___; instead they look like ___." If the
design cannot distinguish them — including ruling out a scale or spatial-autocorrelation artifact —
it does not yet identify the contribution.
Referee pushback → Annals-specific fix
| Likely objection |
Area |
The fix |
| "Your OLS ignores spatial autocorrelation." |
Methods/Human |
Test residual Moran's I; move to a spatial model and report diagnostics. |
| "This is a unit-of-analysis artifact (MAUP)." |
Methods/Nature-Society |
Re-run across areal units/bandwidths; show stability or scope the claim by scale. |
| "Random CV overstates accuracy on spatial data." |
Methods/RS |
Use blocked/spatial CV; report the spatial structure of error. |
| "No independent validation of the classification." |
RS/Physical |
Add a held-out reference sample + area-adjusted accuracy. |
| "Convenience case; what is it a case of?" |
Human/Nature-Society |
State the case-selection logic and the population it represents. |
| "Whose voice / positionality?" |
Human |
Make reflexivity and interpretation-checking explicit. |
Calibration anchors
- Space is a design issue, not a covariate. Dependence, scale, projection, and sampling are decided
in the design, not patched in robustness.
- Each tradition on its own terms. A qualitative design is not weaker for lacking an estimand; it
needs case logic, reflexivity, and disconfirmation criteria instead.
- Mixed means integrated. Two parallel analyses are not mixed methods; specify the linkage.
Anti-patterns
- Ignoring spatial autocorrelation, then reporting OLS SEs as if observations were independent
- No MAUP/scale sensitivity when the result could be a unit-of-analysis artifact
- Classification/prediction with no independent validation, or random CV on spatial data
- Convenience case selection dressed up as theory-driven; positionality omitted in interpretive work
- A nature-society design that never actually links the two strands
Output format
【Mode】spatial-quant / remote-sensing-physical / qualitative / mixed
【Estimand or claim】what is identified/shown
【Spatial integrity】dependence / MAUP-scale / projection / validation handled? [Y/N]
【Rival ruled out】the adjudication sentence (incl. scale/spatial-artifact)
【Robustness】planned checks
【Next】aaag-data-analysis
Supplementary resources
1---2name: aaag-research-design3description: Use when defending the research design of an Annals of the American Association of Geographers manuscript — spatial/quantitative analysis and GIScience, remote-sensing and physical-environmental methods, qualitative human-geography inference, or nature-society mixed methods. The Annals judges each tradition on its own terms. Strengthens the design; it does not write code.4---56# Research Design (aaag-research-design)78The Annals spans four areas and accepts many methodologies, but is demanding about each. The design9must credibly connect the geographic argument (`aaag-theory-building`) to the evidence, and must take10**space and scale seriously** — spatial dependence, the MAUP, projection, and sampling are design11issues, not afterthoughts. This skill is mode-aware: pick the section that matches your work.1213## When to trigger1415- Specifying identification, sampling, case selection, or measurement16- A reviewer questioned spatial autocorrelation, scale/MAUP, edge effects, validation, or a confound17- Justifying why the design adjudicates the rival account from `aaag-literature-positioning`1819## Spatial / quantitative analysis & GIScience20- **Take space seriously.** Test and model **spatial dependence** (Moran's I, spatial lag/error,21 GWR/MGWR where heterogeneity is the point); state how the **MAUP / scale** could change conclusions.22- **Geography of the data.** Document projection/CRS, areal units, edge effects, and the support of23 measurements; spatial sampling and its biases.24- **Inference.** Cluster or use spatial SEs at the right level; for spatial autocorrelation, report25 diagnostics; for prediction, use spatially-aware cross-validation (blocked/spatial CV), not random folds.2627## Remote sensing / physical-environmental28- **Measurement validity.** Sensor/resolution choices, atmospheric/geometric correction, and ground29 truth; quantify accuracy (confusion matrix, kappa/F1, RMSE) with an independent validation sample.30- **Process linkage.** Tie observed pattern to an earth-surface process and its scale; state the31 uncertainty budget end to end.3233## Qualitative / human-geography34- **Case selection** by design logic (typical, extreme, paired, regional contrast) — say what the case35 is a *case of*. Convenience is not a rationale.36- **Positionality, reflexivity, and rigor** appropriate to the method (ethnography, interviews,37 archives, discourse/textual analysis); state how interpretations were checked.38- **Source/field transparency**: plan how fieldnotes, interviews, and archives are documented and cited39 (see `aaag-transparency-and-data`), including consent and geoprivacy.4041## Nature-society / mixed methods42- **Integrate, don't staple.** Specify how the biophysical and social strands inform one another43 (e.g., land-change observation + livelihood interviews), and how convergence/divergence is handled.4445## The adjudication test (Annals-specific)4647For the **single strongest rival explanation**, write: *"If the rival held rather than my argument, the48[spatial pattern / measurements / accounts] would look like ___; instead they look like ___."* If the49design cannot distinguish them — including ruling out a **scale or spatial-autocorrelation artifact** —50it does not yet identify the contribution.5152## Referee pushback → Annals-specific fix5354| Likely objection | Area | The fix |55|------------------|------|---------|56| "Your OLS ignores spatial autocorrelation." | Methods/Human | Test residual Moran's I; move to a spatial model and report diagnostics. |57| "This is a unit-of-analysis artifact (MAUP)." | Methods/Nature-Society | Re-run across areal units/bandwidths; show stability or scope the claim by scale. |58| "Random CV overstates accuracy on spatial data." | Methods/RS | Use blocked/spatial CV; report the spatial structure of error. |59| "No independent validation of the classification." | RS/Physical | Add a held-out reference sample + area-adjusted accuracy. |60| "Convenience case; what is it a case *of*?" | Human/Nature-Society | State the case-selection logic and the population it represents. |61| "Whose voice / positionality?" | Human | Make reflexivity and interpretation-checking explicit. |6263## Calibration anchors6465- **Space is a design issue, not a covariate.** Dependence, scale, projection, and sampling are decided66 in the design, not patched in robustness.67- **Each tradition on its own terms.** A qualitative design is not weaker for lacking an estimand; it68 needs case logic, reflexivity, and disconfirmation criteria instead.69- **Mixed means integrated.** Two parallel analyses are not mixed methods; specify the linkage.7071## Anti-patterns7273- Ignoring spatial autocorrelation, then reporting OLS SEs as if observations were independent74- No MAUP/scale sensitivity when the result could be a unit-of-analysis artifact75- Classification/prediction with no independent validation, or random CV on spatial data76- Convenience case selection dressed up as theory-driven; positionality omitted in interpretive work77- A nature-society design that never actually links the two strands7879## Output format8081```82【Mode】spatial-quant / remote-sensing-physical / qualitative / mixed83【Estimand or claim】what is identified/shown84【Spatial integrity】dependence / MAUP-scale / projection / validation handled? [Y/N]85【Rival ruled out】the adjudication sentence (incl. scale/spatial-artifact)86【Robustness】planned checks87【Next】aaag-data-analysis88```8990## Supplementary resources9192- [`../../resources/external_tools.md`](../../resources/external_tools.md) — spatial-analysis, GIS, remote-sensing, and CAQDAS tooling by area93- [`../../resources/official-source-map.md`](../../resources/official-source-map.md) — areas, scope, and review model
brycewang-stanford