Communicating Model Results Responsibly: Uncertainty, Assumptions, and Trust

Last Updated June 7, 2026

Communicating model results responsibly means explaining what a systems model shows, what it assumes, what it cannot show, and how its outputs should and should not be used. A model result is never just a number, graph, map, ranking, forecast, scenario, dashboard, simulation trace, or optimization recommendation. It is a conditional statement produced under specific assumptions, data limits, model boundaries, parameter choices, validation evidence, uncertainty conditions, and interpretive purposes.

Systems models can improve public reasoning when their results are communicated with clarity, humility, and context. They can reveal feedback loops, delays, accumulations, dependencies, threshold risks, tradeoffs, uncertainty, scenario differences, and intervention consequences that are difficult to see through intuition alone. But model results can also mislead when communicated without caveats. A conditional scenario can become a forecast. A sensitivity test can become a conclusion. A model ranking can become a mandate. A dashboard can look more certain than the evidence permits. A single number can hide uncertainty, assumptions, excluded harms, and distributional effects.

Responsible communication is therefore part of model quality. It is not public relations after the technical work is complete. The way results are explained determines whether the model supports learning, accountability, and better decisions — or whether it creates false precision, hides uncertainty, displaces judgment, and shields institutions from responsibility.

This article examines how to communicate model results responsibly in systems modeling. It covers audiences, uncertainty, assumptions, confidence, scenario framing, visualizations, dashboards, maps, rankings, model cards, decision memos, stakeholder communication, public communication, valid-use statements, misuse warnings, mathematical framing, R and Python workflows, common pitfalls, and practical safeguards for model-based decision-making.

This article covers responsible model communication, uncertainty framing, scenario language, assumption disclosure, confidence statements, visual communication, dashboards, maps, rankings, stakeholder briefings, decision memos, model cards, misuse prevention, public accountability, mathematical framing, R and Python workflows, common pitfalls, and authoritative references.

Why Model Communication Matters

Model communication matters because models influence action. A systems model can shape a budget, policy, infrastructure investment, climate adaptation pathway, public health response, maintenance schedule, risk ranking, resource allocation, organizational strategy, or emergency decision. The people using the model may not have built it. The people affected by the model may not understand its assumptions. The people presenting the model may simplify it. The people making the final decision may remember only the headline result.

That means communication is not a secondary task. It is one of the main pathways through which model quality becomes public consequence. A careful model can be misused if communicated carelessly. A limited model can become dangerous if its limitations are stripped away. A useful scenario analysis can become misleading if one scenario is presented as a forecast. A valid operational model can become harmful if used for policy decisions outside its scope.

The goal is not to make model communication longer. The goal is to make it more truthful, more usable, and less likely to create false confidence.

Back to top ↑

Model Results Are Conditional Statements

A model result is best understood as a conditional statement: under these assumptions, using these data, within this boundary, with this structure, for this purpose, the model produces this result. If those conditions change, the result may change.

This is true even when the result appears exact. A forecast, score, probability, simulation path, optimized allocation, or scenario ranking is not a detached fact. It is produced by a modeling chain. That chain includes the model purpose, system boundary, data sources, parameter estimates, structural assumptions, uncertainty treatment, scenario definitions, validation evidence, and interpretation rules.

Responsible communication keeps the condition attached to the result. It prevents the model output from being detached from the assumptions that give it meaning.

Back to top ↑

Audiences for Model Results

Different audiences need different levels of detail, but no audience should receive misleading certainty. Technical teams may need equations, code, calibration diagnostics, validation results, and sensitivity analysis. Decision-makers may need scenario implications, key assumptions, uncertainty ranges, tradeoffs, and valid-use constraints. Stakeholders may need plain-language explanations of what the model includes, what it excludes, who may be affected, and how results can be challenged.

Responsible communication is audience-aware but not audience-manipulative. It translates complexity without erasing uncertainty, assumptions, or accountability.

Back to top ↑

What Responsible Communication Must Include

A responsible model result should include enough context for users to understand what the model can support and what it cannot support. The required level of detail depends on the model’s stakes. A low-stakes learning model may need a short explanation. A model used for public policy, safety, health, environment, infrastructure, rights, surveillance, or resource allocation needs stronger documentation and communication safeguards.

A model result without these elements may still be technically correct, but it is communicatively incomplete.

Back to top ↑

Communicating Uncertainty

Uncertainty communication is one of the most important parts of responsible model communication. Uncertainty should not be treated as an embarrassment or hidden to make the model sound decisive. In systems modeling, uncertainty is often unavoidable because complex systems involve changing conditions, feedback, human behavior, incomplete data, nonlinear response, ambiguous boundaries, and long time horizons.

Communicating uncertainty does not mean saying “we do not know anything.” It means explaining what is known, what is uncertain, why it is uncertain, how much it matters, and what decisions remain robust despite uncertainty.

Uncertainty communication should be tied to decision relevance. Users need to know not only that uncertainty exists, but which uncertainties could change the conclusion.

Back to top ↑

Communicating Assumptions and Boundaries

Assumptions and boundaries should not be buried in technical appendices alone. The most consequential assumptions should appear wherever results are interpreted. A decision-maker or stakeholder should be able to answer: what must be true for this result to be meaningful?

Boundary communication is especially important because excluded elements can disappear from public reasoning. If a model excludes community disruption, ecological harm, unpaid labor, implementation delays, trust, political feasibility, or long-term maintenance burden, those exclusions must be visible. Otherwise, the model may make a narrow result look like a complete evaluation.

A responsible model briefing should summarize the assumption register and exclusion log in plain language. Users do not need every line of technical documentation in the main text, but they do need the assumptions that can change interpretation.

Back to top ↑

Scenario Results and Forecast Confusion

Scenario results are often misunderstood. A scenario is a structured “what if,” not a prediction. It asks what could happen under stated assumptions. A forecast estimates what is expected to happen. A projection extends assumptions forward. A stress test asks how the system behaves under adverse or extreme conditions.

When scenario language is sloppy, model communication becomes dangerous. A scenario can be mistaken for a forecast, a stress test can be mistaken for expected conditions, and a modeled pathway can be mistaken for a plan. Responsible communication labels each result type clearly.

Every scenario result should answer three questions: what assumptions define the scenario, what output changes under that scenario, and what the scenario should not be interpreted to mean.

Back to top ↑

Confidence, Calibrated Language, and Traceability

Responsible model communication should distinguish confidence from preference. A team may prefer a policy option while having low confidence in parts of the evidence. It may have high confidence in a mechanism but low confidence in parameter values. It may have strong historical validation but weak confidence under future shock conditions.

Calibrated language helps avoid vague claims. Instead of saying “the model proves,” a responsible statement says “the model suggests,” “the model estimates,” “the model is sensitive to,” “the model has high confidence within this range,” or “the model should not be used for this purpose.” Traceability connects each conclusion to assumptions, data, evidence, and validation status.

Traceable communication means users can follow a result back to the model structure, data, assumptions, and review process that produced it.

Back to top ↑

Visualizing Model Results

Visualizations can clarify systems modeling results, but they can also create false confidence. A graph, map, dashboard, or flow diagram may look authoritative even when uncertainty is high. Responsible visualization should make uncertainty, scale, boundary, and context visible rather than hiding them.

A responsible visualization should not merely be attractive. It should help users understand what the model result means and what it does not mean.

Back to top ↑

Dashboards, Maps, and Rankings

Dashboards, maps, and rankings require special care because they compress complexity into actionable displays. They are useful precisely because they simplify, but that simplification can create overconfidence.

A dashboard can make data look current even when inputs are stale. A map can make uncertainty look geographically precise. A ranking can make small score differences look meaningful. A red-yellow-green status indicator can hide assumptions, thresholds, and exceptions. A composite score can hide the weights and proxies that created it.

Responsible interface design keeps caveats attached to outputs. Caveats should not disappear when users hover, filter, export, screenshot, or present model results.

Back to top ↑

Distributional and Equity Communication

Aggregate model results can hide who benefits and who bears burden. A model may show that a policy improves average reliability, reduces total cost, or increases system efficiency while worsening conditions for specific groups or places. Responsible communication must report distributional effects when model results affect access, risk, resources, exposure, rights, health, environment, or public services.

Distributional communication is not optional when model results may affect unequal systems. Average performance is not the same as fair performance.

Back to top ↑

Decision Memos, Model Cards, and Documentation

Responsible communication benefits from structured documentation. A model result should travel with a short, durable explanation of its purpose, data, assumptions, validation, uncertainty, intended use, limitations, and misuse risks. This documentation can take different forms depending on the audience and stakes.

The most important documentation should be short enough to travel with the result. Long technical reports matter, but they do not protect users if key caveats vanish from the materials people actually read.

Back to top ↑

Public Communication and Accountability

Public model communication should be especially careful because public audiences may not have access to the technical context. A public-facing result should not imply that the model decided, that uncertainty is absent, or that values were purely technical. It should explain why the model was used, what role it played, what assumptions matter, what the model cannot answer, and who remains accountable for the final decision.

Public communication should not use model complexity to close debate. It should use model clarity to improve debate.

Back to top ↑

Misuse Warnings and Valid-Use Statements

Every consequential model should have a valid-use statement and misuse warnings. A valid-use statement explains what the model is appropriate for. A misuse warning explains what the model should not be used to do.

This is especially important when model outputs are likely to be reused, summarized, exported, shared, or embedded in decision workflows. Without explicit use limits, a model built for scenario exploration may be used for prediction. A screening model may be used for denial of services. A planning model may be used for real-time control. A dashboard may be used as if it were a complete representation of reality.

Valid-use statements protect both decision-makers and affected communities. They make it harder for model outputs to drift into uses the model cannot support.

Back to top ↑

Communication Across Modeling Paradigms

Different modeling paradigms require different communication safeguards. A system dynamics model needs clear explanation of stocks, flows, feedback loops, delays, and aggregation. An agent-based model needs clear explanation of agent rules, heterogeneity, interaction, and calibration. A network model needs clear explanation of nodes, edges, weights, dependency, and missing ties. An AI-assisted model needs clear explanation of training data, performance limits, bias, intended use, and interpretability.

The communication method should match the modeling method. But every method requires the same ethical discipline: assumptions, uncertainty, limits, and accountability must travel with results.

Back to top ↑

Mathematical Lens: Result, Uncertainty, Confidence, and Communication Risk

A model result can be represented as a conditional output:

Uncertainty can be represented as a range rather than a single value:

Communication risk increases when apparent precision exceeds justified precision:

Sensitivity can be represented as the change in output caused by a change in an assumption or parameter:

Decision relevance can combine effect size and uncertainty:

A model communication score can combine clarity, uncertainty disclosure, assumption disclosure, and misuse safeguards:

These formulas do not automate communication judgment. They show why responsible communication is part of model interpretation, not decoration.

Back to top ↑

Back to top ↑

Strengths and Limitations

Responsible communication improves model usefulness because it helps users interpret results correctly. It does not weaken the model by adding caveats. It strengthens the model by preventing overclaiming, false precision, misuse, and loss of accountability.

The limitation is that communication alone cannot fix a flawed model. If the model boundary is inappropriate, data are biased, assumptions are weak, or validation is inadequate, communication should reveal those problems — not cover them.

Back to top ↑

R Workflow: Model Communication Risk and Uncertainty Summary

The R workflow below uses base R only. It creates synthetic model results, scores communication quality, identifies false-precision risk, summarizes uncertainty disclosure, and exports validation checks. It is designed as a lightweight companion workflow for model result communication review.

# communicating_model_results_responsibly_workflow.R
# Base R workflow:
# model communication risk, uncertainty disclosure, and valid-use checks.
#
# Suggested repository placement:
# articles/communicating-model-results-responsibly/r/communicating_model_results_responsibly_workflow.R

args <- commandArgs(trailingOnly = FALSE)
file_arg <- grep("^--file=", args, value = TRUE)

if (length(file_arg) > 0) {
  script_path <- normalizePath(sub("^--file=", "", file_arg[1]), mustWork = TRUE)
  article_root <- normalizePath(file.path(dirname(script_path), ".."), mustWork = TRUE)
} else {
  article_root <- normalizePath(getwd(), mustWork = TRUE)
}

tables_dir <- file.path(article_root, "outputs", "tables")
figures_dir <- file.path(article_root, "outputs", "figures")

dir.create(tables_dir, recursive = TRUE, showWarnings = FALSE)
dir.create(figures_dir, recursive = TRUE, showWarnings = FALSE)

model_results <- data.frame(
  result_id = c("R1", "R2", "R3", "R4", "R5"),
  result_type = c("scenario", "forecast", "ranking", "map", "optimization"),
  point_estimate = c(0.72, 12000, 0.84, 0.67, 0.91),
  lower_bound = c(0.55, 9000, 0.75, 0.40, 0.80),
  upper_bound = c(0.88, 16000, 0.89, 0.82, 0.96),
  assumption_disclosure = c(0.80, 0.60, 0.70, 0.45, 0.65),
  uncertainty_disclosure = c(0.85, 0.75, 0.55, 0.40, 0.60),
  boundary_disclosure = c(0.70, 0.55, 0.65, 0.50, 0.60),
  misuse_warning = c(0.75, 0.60, 0.45, 0.40, 0.55)
)

model_results$uncertainty_width <- model_results$upper_bound - model_results$lower_bound

model_results$communication_quality_score <-
  0.30 * model_results$assumption_disclosure +
  0.30 * model_results$uncertainty_disclosure +
  0.20 * model_results$boundary_disclosure +
  0.20 * model_results$misuse_warning

model_results$false_precision_risk <-
  ifelse(
    model_results$uncertainty_disclosure < 0.60 & model_results$uncertainty_width > 0.20,
    "high",
    ifelse(model_results$uncertainty_disclosure < 0.70, "moderate", "lower")
  )

communication_controls <- data.frame(
  control_id = c("C1", "C2", "C3", "C4", "C5", "C6"),
  control = c(
    "state_model_purpose",
    "show_uncertainty_range",
    "summarize_key_assumptions",
    "label_scenarios_as_conditional",
    "publish_valid_use_statement",
    "attach_misuse_warning"
  ),
  present = c(TRUE, TRUE, TRUE, TRUE, FALSE, FALSE),
  risk_if_absent = c(
    "Users may not understand what question the model answers.",
    "Point estimates may appear certain.",
    "Conditional results may appear unconditional.",
    "Scenarios may be mistaken for forecasts.",
    "Model may be used beyond scope.",
    "Likely misuse may go unchallenged."
  )
)

validation_checks <- data.frame(
  check = c(
    "model_results_created",
    "communication_scores_between_zero_and_one",
    "uncertainty_widths_nonnegative",
    "communication_controls_created"
  ),
  passed = c(
    nrow(model_results) > 0,
    all(model_results$communication_quality_score >= 0 & model_results$communication_quality_score <= 1),
    all(model_results$uncertainty_width >= 0),
    nrow(communication_controls) > 0
  )
)

write.csv(
  model_results,
  file.path(tables_dir, "r_model_result_communication_diagnostics.csv"),
  row.names = FALSE
)

write.csv(
  communication_controls,
  file.path(tables_dir, "r_model_communication_controls.csv"),
  row.names = FALSE
)

write.csv(
  validation_checks,
  file.path(tables_dir, "r_model_communication_validation_checks.csv"),
  row.names = FALSE
)

png(file.path(figures_dir, "r_communication_quality_scores.png"), width = 1000, height = 700)
barplot(
  model_results$communication_quality_score,
  names.arg = model_results$result_id,
  ylab = "Communication Quality Score",
  xlab = "Result ID",
  main = "Model Result Communication Quality"
)
grid()
dev.off()

print(model_results)
print(communication_controls)
print(validation_checks)
cat("R communicating model results responsibly workflow complete.\n")

This workflow treats model communication as a reviewable artifact. It does not determine whether a result is true. It checks whether the result is being communicated with enough context to reduce avoidable misuse.

Back to top ↑

Python Workflow: Model Result Briefing and Communication Controls

The Python workflow below uses only the standard library. It creates model result summaries, uncertainty ranges, communication quality scores, false-precision risk labels, valid-use statements, misuse warnings, and validation checks.

#!/usr/bin/env python3
"""
Communicating model results responsibly workflow.

Dependency-light workflow demonstrating:

1. Model result briefing tables
2. Uncertainty range calculation
3. Communication quality scoring
4. False precision risk labels
5. Valid-use and misuse controls
6. Validation checks

All data are synthetic.
"""

from __future__ import annotations

from pathlib import Path
import csv


ARTICLE_ROOT = Path(__file__).resolve().parents[1]
TABLES = ARTICLE_ROOT / "outputs" / "tables"


def write_csv(path: Path, rows: list[dict[str, object]]) -> None:
    path.parent.mkdir(parents=True, exist_ok=True)
    if not rows:
        raise ValueError(f"No rows to write: {path}")

    fieldnames: list[str] = []
    for row in rows:
        for key in row:
            if key not in fieldnames:
                fieldnames.append(key)

    with path.open("w", newline="", encoding="utf-8") as handle:
        writer = csv.DictWriter(handle, fieldnames=fieldnames, extrasaction="ignore")
        writer.writeheader()
        writer.writerows(rows)


def false_precision_label(uncertainty_disclosure: float, uncertainty_width: float) -> str:
    if uncertainty_disclosure < 0.60 and uncertainty_width > 0.20:
        return "high_false_precision_risk"
    if uncertainty_disclosure < 0.70:
        return "moderate_false_precision_risk"
    return "lower_false_precision_risk"


def main() -> None:
    model_results = [
        {
            "result_id": "R1",
            "result_type": "scenario",
            "point_estimate": 0.72,
            "lower_bound": 0.55,
            "upper_bound": 0.88,
            "assumption_disclosure": 0.80,
            "uncertainty_disclosure": 0.85,
            "boundary_disclosure": 0.70,
            "misuse_warning": 0.75,
            "plain_language_result": "Scenario A improves modeled reliability under stated assumptions.",
        },
        {
            "result_id": "R2",
            "result_type": "forecast",
            "point_estimate": 12000,
            "lower_bound": 9000,
            "upper_bound": 16000,
            "assumption_disclosure": 0.60,
            "uncertainty_disclosure": 0.75,
            "boundary_disclosure": 0.55,
            "misuse_warning": 0.60,
            "plain_language_result": "Demand is forecast to rise, but the plausible range is wide.",
        },
        {
            "result_id": "R3",
            "result_type": "ranking",
            "point_estimate": 0.84,
            "lower_bound": 0.75,
            "upper_bound": 0.89,
            "assumption_disclosure": 0.70,
            "uncertainty_disclosure": 0.55,
            "boundary_disclosure": 0.65,
            "misuse_warning": 0.45,
            "plain_language_result": "Option B ranks highest under selected weights.",
        },
        {
            "result_id": "R4",
            "result_type": "map",
            "point_estimate": 0.67,
            "lower_bound": 0.40,
            "upper_bound": 0.82,
            "assumption_disclosure": 0.45,
            "uncertainty_disclosure": 0.40,
            "boundary_disclosure": 0.50,
            "misuse_warning": 0.40,
            "plain_language_result": "Mapped exposure is uneven, but data coverage varies by location.",
        },
        {
            "result_id": "R5",
            "result_type": "optimization",
            "point_estimate": 0.91,
            "lower_bound": 0.80,
            "upper_bound": 0.96,
            "assumption_disclosure": 0.65,
            "uncertainty_disclosure": 0.60,
            "boundary_disclosure": 0.60,
            "misuse_warning": 0.55,
            "plain_language_result": "The optimized option performs best under the stated objective function.",
        },
    ]

    for row in model_results:
        uncertainty_width = float(row["upper_bound"]) - float(row["lower_bound"])
        communication_quality_score = (
            0.30 * float(row["assumption_disclosure"])
            + 0.30 * float(row["uncertainty_disclosure"])
            + 0.20 * float(row["boundary_disclosure"])
            + 0.20 * float(row["misuse_warning"])
        )

        row["uncertainty_width"] = round(uncertainty_width, 6)
        row["communication_quality_score"] = round(communication_quality_score, 6)
        row["false_precision_risk"] = false_precision_label(
            float(row["uncertainty_disclosure"]),
            uncertainty_width,
        )

    communication_controls = [
        {
            "control_id": "C1",
            "control": "state_model_purpose",
            "present": True,
            "risk_if_absent": "Users may not understand what question the model answers.",
        },
        {
            "control_id": "C2",
            "control": "show_uncertainty_range",
            "present": True,
            "risk_if_absent": "Point estimates may appear certain.",
        },
        {
            "control_id": "C3",
            "control": "summarize_key_assumptions",
            "present": True,
            "risk_if_absent": "Conditional results may appear unconditional.",
        },
        {
            "control_id": "C4",
            "control": "label_scenarios_as_conditional",
            "present": True,
            "risk_if_absent": "Scenarios may be mistaken for forecasts.",
        },
        {
            "control_id": "C5",
            "control": "publish_valid_use_statement",
            "present": False,
            "risk_if_absent": "Model may be used beyond scope.",
        },
        {
            "control_id": "C6",
            "control": "attach_misuse_warning",
            "present": False,
            "risk_if_absent": "Likely misuse may go unchallenged.",
        },
    ]

    valid_use_register = [
        {
            "result_type": "scenario",
            "valid_use": "Compare conditional outcomes under stated assumptions.",
            "misuse_warning": "Do not present scenarios as predictions.",
        },
        {
            "result_type": "forecast",
            "valid_use": "Estimate expected conditions within the validated domain.",
            "misuse_warning": "Do not ignore uncertainty ranges or structural change.",
        },
        {
            "result_type": "ranking",
            "valid_use": "Support prioritization for review.",
            "misuse_warning": "Do not treat small score differences as decisive without sensitivity analysis.",
        },
        {
            "result_type": "map",
            "valid_use": "Show spatial patterns at the stated resolution.",
            "misuse_warning": "Do not infer precision below the data resolution.",
        },
        {
            "result_type": "optimization",
            "valid_use": "Compare options under stated objectives and constraints.",
            "misuse_warning": "Do not treat the objective function as a complete value judgment.",
        },
    ]

    validation_rows = [
        {
            "check": "model_results_created",
            "passed": len(model_results) > 0,
            "value": len(model_results),
        },
        {
            "check": "communication_scores_between_zero_and_one",
            "passed": all(0 <= float(row["communication_quality_score"]) <= 1 for row in model_results),
            "value": "all_scores_checked",
        },
        {
            "check": "uncertainty_widths_nonnegative",
            "passed": all(float(row["uncertainty_width"]) >= 0 for row in model_results),
            "value": "all_widths_checked",
        },
        {
            "check": "communication_controls_created",
            "passed": len(communication_controls) > 0,
            "value": len(communication_controls),
        },
    ]

    write_csv(TABLES / "python_model_result_communication_diagnostics.csv", model_results)
    write_csv(TABLES / "python_model_communication_controls.csv", communication_controls)
    write_csv(TABLES / "python_model_valid_use_register.csv", valid_use_register)
    write_csv(TABLES / "python_model_communication_validation_checks.csv", validation_rows)

    print("Communicating model results responsibly workflow complete.")
    print(TABLES / "python_model_result_communication_diagnostics.csv")


if __name__ == "__main__":
    main()

This workflow demonstrates how a model result can be packaged with uncertainty, assumptions, valid-use limits, and misuse warnings. The communication score is not a substitute for judgment; it is a prompt for review.

Back to top ↑

GitHub Repository

Back to top ↑

Common Pitfalls

Model communication often fails in predictable ways. The most common problem is not that the model is too complex. It is that the communication removes the context needed to interpret the model responsibly.

The central correction is simple: keep model meaning attached to model outputs.

Back to top ↑

Conclusion

Communicating model results responsibly is part of responsible systems modeling. A model result is not self-explanatory. It must be interpreted through assumptions, data, boundaries, uncertainty, validation, scenarios, values, and intended use. When those conditions are communicated clearly, models can improve reasoning. When they are hidden, models can mislead.

Responsible communication does not mean overwhelming users with technical detail. It means giving each audience enough information to understand what the model result means, how confident they should be, what assumptions matter, what uncertainty remains, what consequences are included or excluded, and what uses would be inappropriate.

The strongest model communication is conditional, traceable, calibrated, visualized honestly, and accountable. It says what the model shows without pretending the model shows everything. It communicates uncertainty without paralysis. It supports decision-making without displacing judgment. It makes public reasoning more informed rather than more opaque.

The responsibility of the modeler does not end when the simulation runs. It continues until the result is understood within its proper scope.

Back to top ↑

Related Articles

• Systems Modeling
• What Is Systems Modeling?
• Uncertainty and Model Interpretation
• Model Assumptions and Boundary Judgment
• When Systems Models Clarify and When They Distort
• Ethics, Power, and Systems Modeling
• Calibration and Validation of Models
• Sensitivity Analysis in Systems Models
• Participatory Modeling and Stakeholder Systems
• Decision Science
• Public Policy Modeling
• Future Directions in Systems Modeling

Further Reading

• Saltelli, A., et al. (2020) ‘Five ways to ensure that models serve society: a manifesto’, Nature, 582, pp. 482–484. Available at: https://www.nature.com/articles/d41586-020-01812-9.
• National Academies of Sciences, Engineering, and Medicine. (2017) Communicating Science Effectively: A Research Agenda. Washington, DC: National Academies Press. Available at: https://www.nationalacademies.org/publications/23674/communicating-science-effectively-a-research-agenda.
• Fischhoff, B. (2006) Communicating Uncertainty. National Academies. Available at: https://sites.nationalacademies.org/cs/groups/dbassesite/documents/webpage/dbasse_070995.pdf.
• Mastrandrea, M.D., et al. (2010) Guidance Note for Lead Authors of the IPCC Fifth Assessment Report on Consistent Treatment of Uncertainties. Available at: https://www.ipcc.ch/site/assets/uploads/2018/05/uncertainty-guidance-note.pdf.
• Mitchell, M., et al. (2019) ‘Model Cards for Model Reporting’, Proceedings of the Conference on Fairness, Accountability, and Transparency. Available at: https://arxiv.org/abs/1810.03993.
• Pushkarna, M., Zaldivar, A. and Kjartansson, O. (2022) ‘Data Cards: Purposeful and Transparent Dataset Documentation for Responsible AI’, FAccT ’22. Available at: https://arxiv.org/abs/2204.01075.
• National Research Council. (2012) Assessing the Reliability of Complex Models: Mathematical and Statistical Foundations of Verification, Validation, and Uncertainty Quantification. Washington, DC: National Academies Press. Available at: https://www.nationalacademies.org/publications/13395/assessing-the-reliability-of-complex-models.
• Oreskes, N., Shrader-Frechette, K. and Belitz, K. (1994) ‘Verification, validation, and confirmation of numerical models in the Earth sciences’, Science, 263(5147), pp. 641–646.
• Spiegelhalter, D., Pearson, M. and Short, I. (2011) ‘Visualizing uncertainty about the future’, Science, 333(6048), pp. 1393–1400.
• Hullman, J. (2020) ‘Why Authors Don’t Visualize Uncertainty’, IEEE Transactions on Visualization and Computer Graphics, 26(1), pp. 130–139.
• Saltelli, A., Tarantola, S., Campolongo, F. and Ratto, M. (2008) Global Sensitivity Analysis: The Primer. Chichester: Wiley.
• Sterman, J.D. (2000) Business Dynamics: Systems Thinking and Modeling for a Complex World. Boston: Irwin/McGraw-Hill.
• Meadows, D.H. (2008) Thinking in Systems: A Primer. White River Junction, VT: Chelsea Green.

Back to top ↑

References

• Fischhoff, B. (2006) Communicating Uncertainty. National Academies. Available at: https://sites.nationalacademies.org/cs/groups/dbassesite/documents/webpage/dbasse_070995.pdf.
• Hullman, J. (2020) ‘Why Authors Don’t Visualize Uncertainty’, IEEE Transactions on Visualization and Computer Graphics, 26(1), pp. 130–139.
• Mastrandrea, M.D., et al. (2010) Guidance Note for Lead Authors of the IPCC Fifth Assessment Report on Consistent Treatment of Uncertainties. Available at: https://www.ipcc.ch/site/assets/uploads/2018/05/uncertainty-guidance-note.pdf.
• Meadows, D.H. (2008) Thinking in Systems: A Primer. White River Junction, VT: Chelsea Green.
• Mitchell, M., et al. (2019) ‘Model Cards for Model Reporting’, Proceedings of the Conference on Fairness, Accountability, and Transparency. Available at: https://arxiv.org/abs/1810.03993.
• National Academies of Sciences, Engineering, and Medicine. (2017) Communicating Science Effectively: A Research Agenda. Washington, DC: National Academies Press. Available at: https://www.nationalacademies.org/publications/23674/communicating-science-effectively-a-research-agenda.
• National Research Council. (2012) Assessing the Reliability of Complex Models: Mathematical and Statistical Foundations of Verification, Validation, and Uncertainty Quantification. Washington, DC: National Academies Press. Available at: https://www.nationalacademies.org/publications/13395/assessing-the-reliability-of-complex-models.
• Oreskes, N., Shrader-Frechette, K. and Belitz, K. (1994) ‘Verification, validation, and confirmation of numerical models in the Earth sciences’, Science, 263(5147), pp. 641–646.
• Pushkarna, M., Zaldivar, A. and Kjartansson, O. (2022) ‘Data Cards: Purposeful and Transparent Dataset Documentation for Responsible AI’, FAccT ’22. Available at: https://arxiv.org/abs/2204.01075.
• Saltelli, A., et al. (2020) ‘Five ways to ensure that models serve society: a manifesto’, Nature, 582, pp. 482–484. Available at: https://www.nature.com/articles/d41586-020-01812-9.
• Saltelli, A., Tarantola, S., Campolongo, F. and Ratto, M. (2008) Global Sensitivity Analysis: The Primer. Chichester: Wiley.
• Spiegelhalter, D., Pearson, M. and Short, I. (2011) ‘Visualizing uncertainty about the future’, Science, 333(6048), pp. 1393–1400.
• Sterman, J.D. (2000) Business Dynamics: Systems Thinking and Modeling for a Complex World. Boston: Irwin/McGraw-Hill.

Back to top ↑