Integration and Implementation Sciences (i2S) · Published 2014-10-08

Models that Matter – System Dynamics Applications with Impact by George Richardson

privateReview: unreviewedConfidence: high
system dynamicsFeedback loopsurban dynamicsproject managementepidemiologysystem-dynamicsfeedback-loopsstock-and-flowcompensating-feedbackendogenous-causalityundiscovered-reworkevaluation-horizoncontrol-loop-oscillation
Open on YouTube ↗

Summary

Overview

  • Speaker: George P. Richardson
  • Channel: Integration and Implementation Sciences (i2S)
  • Main topic: System dynamics modeling and its real-world policy impact
  • Purpose: To demonstrate how system dynamics models generate real-world policy impact through four core application domains and to outline what makes modeling impactful. George P. Richardson discusses system dynamics applications across commodity cycles, project dynamics, urban dynamics, and infectious disease control (polio eradication). He explores what creates impact in system dynamics modeling, emphasizing the importance of significant problem definitions, clear dynamic problem formulation, stakeholder involvement, multiple perspectives, requisite modeling, and effective sensitive communication.

Topic Map

Commodity Cycles

  • Explanation: Analysis of commodity price instabilities and market oscillations worldwide using system dynamics models.
  • Key claims:
    • Commodity markets exhibit inherent instabilities and price oscillations.
    • Simple system dynamics models with inventory, production capacity, and delays can reproduce oscillatory behavior.
    • Random disturbances (noise) can cause a system to oscillate at its natural frequency.
  • Examples:
    • Peanut farming and price volatility in Gambia
    • Artisanal mining and mineral compounds risk in the Philippines
    • Coffee price fluctuations between 1982 and 2012
  • Terminology:
    • commodity cycles
    • inventory
    • production capacity
    • availability loop
    • capacity adjustment delay
  • Why it matters: Understanding commodity cycle dynamics allows policymakers to design interventions that stabilize volatile global markets.

Project Dynamics & Cost Overruns

  • Explanation: Examination of why major projects experience severe overruns in cost, schedule, and materials.
  • Key claims:
    • Cost overruns are a major problem in large projects across many countries.
    • Undiscovered rework is central to the management of complex projects.
    • Project dynamics emerge endogenously from interactions between workforce, schedule, tasks, and perceived progress.
  • Examples:
    • Sydney Olympic Park and Sydney Opera House overruns
    • Litton industry ship building litigation with the US Navy
    • Channel tunnel litigation
  • Terminology:
    • project dynamics
    • undiscovered rework
    • perceived progress
    • real progress
    • workforce plateau
  • Why it matters: Modeling project dynamics helps managers anticipate overruns and avoid pitfalls like halting hiring when projects plateau prematurely.

Urban Dynamics

  • Explanation: Study of urban growth, stagnation, and decay modeled after Forrester's work on city dynamics.
  • Key claims:
    • Major cities exhibit similar lifecycle patterns of growth, plateauing, and decline.
    • Urban decay can be modeled endogenously using housing, business structures, and population interactions.
    • Policies like building low-cost housing or jobs programs can have unintended long-term compensating feedback effects.
  • Examples:
    • Population curves of New York, Chicago, and Philadelphia
    • Urban decay in Camden, New Jersey
    • Urban1 model simulations of housing and employment
  • Terminology:
    • urban dynamics
    • urban decay
    • labor-to-job ratio
    • housing density
    • compensating feedback
  • Why it matters: Recognizing endogenous urban dynamics prevents policymakers from implementing well-intentioned policies that backfire due to system compensation.

Infectious Diseases & Polio Eradication

  • Explanation: Modeling the spread of diseases, quarantine efficacy, and the global economic and policy case for polio eradication.
  • Key claims:
    • Infectious disease spread can be modeled using susceptible, sick, and recovered stocks.
    • System dynamics models played a crucial role in convincing the WHO to shift from outbreak containment to global eradication of polio.
    • Decision risk communication models help bridge rigorous modeling and high-level decision-making.
  • Examples:
    • Influenza and childhood diseases modeling
    • Polio vaccination campaigns in India supported by Gates Foundation
    • Kimberly Thompson's work on economic analysis of poliomyelitis eradication
  • Terminology:
    • epidemiology
    • susceptible population
    • quarantine
    • inoculation
    • stock and flow
  • Why it matters: System dynamics helps public health organizations evaluate policies and shift global strategies from containment to permanent eradication.

Key Points

Bath-tub Thinking

  • Explanation: Visualizing inventory and commodity flows using the mental model of a bathtub with inflow pipes and outflow pipes.
  • Evidence: Described during the commodity model explanation where inventory acts as a tub with producing inflows and shipping outflows.
  • Practical implication: Helps modelers intuitively grasp stock-and-flow structures in complex systems.

Endogenous Explanations

  • Explanation: Deriving system behaviors from internal structure and interactions rather than external shocks.
  • Evidence: Forrester's urban dynamics model derived city growth and decay strictly from internal city policies.
  • Practical implication: Focuses problem-solving efforts on structural changes rather than blaming external factors.

Compensating Feedback

  • Explanation: Natural feedback systems in social and economic structures that counteract single-point policy interventions.
  • Evidence: Demonstrated when a successful jobs program temporarily helps a city, but long-term feedback loops negate the effect.
  • Practical implication: Warns policymakers that short-term benefits can disappear if feedback loops are ignored.

Frameworks, Models & Processes

Simple Commodity Model

  • How it works: Connects production capacity, inventory, and ordering demand in a feedback loop to simulate market price cycles.
  • Components:
    • Inventory stock
    • Producing flow
    • Shipping flow
    • Production capacity
    • Availability loop
  • When to use: When studying market instabilities and price oscillations in commodities.

Project Dynamics Model

  • How it works: Tracks workforce hiring, task completion, perceived progress versus real progress, and undiscovered rework.
  • Components:
    • Workforce stock
    • Task completion
    • Perceived progress
    • Undiscovered rework
  • When to use: When managing complex engineering or construction projects prone to schedule overruns.

Epidemic Model with Quarantine and Inoculation

  • How it works: Models population transitions between susceptible, sick, symptomatic, quarantined, and recovered states.
  • Components:
    • Susceptible stock
    • Sick stock
    • Recovered stock
    • Quarantine rate
    • Inoculation rate
  • When to use: When studying the spread and containment of infectious diseases.

Examples & Case Studies

Litton Industry sued the US Navy for half a billion dollars due to shipbuilding cost overruns.

  • Illustrates: The practical application of system dynamics models in litigation and adjudication of project overruns.
  • Lesson: System dynamics models can serve as objective frameworks for resolving complex financial disputes in large projects.

Kimberly Thompson and colleagues used system dynamics models to convince the WHO to change its polio policy from containment to eradication.

  • Illustrates: How modeling can successfully influence global health policy.
  • Lesson: Rigorous modeling combined with effective stakeholder communication can shift global health strategies.

Actionable Takeaways

  • Immediate:
    • Adopt stock-and-flow bathtub thinking for system analysis.
    • Focus on internal structure over external blame when diagnosing problems.
  • Strategic:
    • Involve stakeholders early in the modeling process to build trust and ownership.
    • Build small, understandable models for learning and large models for confidence.
    • Account for compensating feedback when designing public policies.
  • Questions to investigate:
    • What are the key feedback loops driving oscillations in our target system?
    • How do stakeholder perspectives influence problem definition?

Claims Worth Verifying

  • Denis Meadows wrote his dissertation on commodity cycles in the early 1970s and is a lead author of Limits to Growth. (historical fact)
  • Kimberly Thompson won the Forrester Award from the System Dynamics Society in 2008. (award record)

Notable Quotes

"Stand up so you all can see me, speak up so you all can hear me, and shut up so you all will like me." (at 2:06) "It's bathtub thinking. That's exactly what it is." (at 6:51) "There is no magic from the outside, it's endogenous." (at 21:41)

Compressed Summary

  • System dynamics explains complex behaviors through internal feedback structures.
  • Four major application areas: commodity cycles, project dynamics, urban dynamics, and infectious diseases.
  • Model impact requires significant problems, stakeholder involvement, and requisite modeling.
  • Small models are powerful for learning and understanding, while large models build confidence.
  • Keywords: system dynamics, feedback loops, urban dynamics, project management, epidemiology
  • Core insight: System dynamics modeling achieves real-world impact by capturing endogenous feedback structures and engaging stakeholders in understanding complex, problem-driven systems.

Core insights

7
Mechanismmedium noveltymoderate evidence

Oscillatory behavior in a system can be triggered by random noise rather than structural forcing — noise merely excites the system's natural frequency, so stabilizing interventions must target the structure producing that frequency, not the disturbances.

Why it matters

Teams routinely attribute production instability (latency spikes, retry storms, autoscaler flapping) to external traffic noise and respond with smoothing/damping at the input. If the oscillation frequency is endogenous, input smoothing will not fix it and can mask the real control loop that needs retuning.

Generalization

Any closed-loop control system with delays (capacity adjustment, retries, caches, autoscalers, scheduling) will ring at a natural frequency when perturbed; diagnose the loop, not the perturbation.

Random disturbances (noise) can cause a system to oscillate at its natural frequency.
Open source video
Mechanismhigh noveltystrong evidence

'Undiscovered rework' is the central hidden variable in complex project dynamics: work appears complete under perceived progress but is not, so hiring decisions keyed to perceived progress plateau prematurely while real progress lags.

Why it matters

Directly maps to agentic/software systems where progress signals are proxies (tests passing, tasks marked done, LLM self-reported completion). If the proxy diverges from real completion, control decisions (add workers/agents, increase parallelism) are systematically wrong and generate late-stage rework cascades.

Generalization

Any system that controls effort allocation using a self-reported or partially verified progress metric will accumulate hidden rework; instrument real vs perceived progress separately.

Undiscovered rework is central to the management of complex projects.
Open source video
Project dynamics emerge endogenously from interactions between workforce, schedule, tasks, and perceived progress.
Open source video
halting hiring when projects plateau prematurely
Open source video
Mental Modelmedium noveltymoderate evidence

Endogenous explanation — deriving behavior strictly from internal structure and policy rather than external shocks — is presented as the modeling discipline that redirects problem-solving toward structural change instead of blaming outside factors.

Why it matters

For AI system reliability work, this is the argument for modeling failure as produced by your own architecture (retry policies, tool contracts, context assembly) rather than attributing it to model variance or user inputs; it changes what you fix and what you measure.

Generalization

When a system's behavior is reproducible from its own internal loops, external-cause explanations are unfalsifiable and unactionable; build the model that closes the loop internally.

Deriving system behaviors from internal structure and interactions rather than external shocks.
Open source video
Forrester's urban dynamics model derived city growth and decay strictly from internal city policies.
Open source video
Failure Modemedium noveltymoderate evidence

Compensating feedback: single-point policy interventions (jobs programs, low-cost housing) are neutralized long-term by the system's own feedback loops, so short-term measured benefit is an unreliable predictor of steady-state outcome.

Why it matters

Directly analogous to interventions in engineered systems: adding capacity, raising timeouts, adding agents, or lowering thresholds often shows immediate improvement in short-horizon metrics while the compensating loop (cache pressure, coordination overhead, contention) erases it later. Evaluating only at short horizons systematically selects for interventions that fail.

Generalization

Benchmark and evaluation horizons must exceed the timescale of the slowest compensating feedback loop, or measurements will invert the true ranking of interventions.

Policies like building low-cost housing or jobs programs can have unintended long-term compensating feedback effects.
Open source video
a successful jobs program temporarily helps a city, but long-term feedback loops negate the effect
Open source video
Practicemedium noveltymoderate evidence

Impactful modeling is framed as a set of non-technical design constraints — significant problem definition, clear dynamic problem formulation, stakeholder involvement, multiple perspectives, requisite modeling, and sensitive communication — rather than model fidelity.

Why it matters

Suggests that for agentic systems, adoption/reliability outcomes may be gated more by problem formulation and audience-calibrated communication of model output (decision risk communication) than by the capability of the underlying model.

Generalization

Model output that cannot be trusted or interpreted by the decision-maker does not change decisions; design the interface between model and decision as a first-class artifact.

emphasizing the importance of significant problem definitions, clear dynamic problem formulation, stakeholder involvement, multiple perspectives, requisite modeling, and effective sensitive communication
Open source video
Decision risk communication models help bridge rigorous modeling and high-level decision-making.
Open source video
Architecturemedium noveltymoderate evidence

A minimal stock-and-flow vocabulary (inventory, capacity, delays) or even a physical mental model ('bathtub thinking') is sufficient to reproduce real oscillatory behavior in complex markets.

Why it matters

Argues for small, structurally faithful abstractions over high-dimensional fitted models when the goal is to explain and intervene rather than predict: the abstraction is what makes the loop legible and the intervention point identifiable.

Generalization

Prefer minimal causal-structure models that can be interrogated and modified over black-box models that fit better but hide the feedback topology.

Simple system dynamics models with inventory, production capacity, and delays can reproduce oscillatory behavior.
Open source video
Visualizing inventory and commodity flows using the mental model of a bathtub with inflow pipes and outflow pipes.
Open source video
Empirical Resultmedium noveltymoderate evidence

Modeling output was used as evidence in high-stakes adjudication (Litton v. US Navy shipbuilding overruns, Channel tunnel litigation) and to shift a global institution's strategy from outbreak containment to eradication.

Why it matters

Shows a mechanism/model achieving causal force on real resource allocation and policy at the highest level, which is the bar for arguing that a model is decision-grade rather than descriptive.

Generalization

The strongest validation of a system model is a change in a consequential external decision it caused, not internal fit metrics.

Litton Industry sued the US Navy for half a billion dollars due to shipbuilding cost overruns.
Open source video
System dynamics models played a crucial role in convincing the WHO to shift from outbreak containment to global eradication of polio.
Open source video

Deep dives

4

Online detection of undiscovered rework in agentic and software delivery pipelines

Research question

Can divergence between a self-reported progress signal (tasks marked done, tests passing, agent-declared completion) and an independently verified completion signal be measured online, and does the divergence rate at time T predict rework volume at time T+n?

Why

If control decisions (add agents, raise parallelism, halt hiring) are keyed to perceived progress, hidden rework accumulates invisibly and surfaces only as late-stage overrun. A dual-signal instrument turns 'undiscovered' rework into a measurable stock with a leading indicator, which is the precondition for any intervention.

Undiscovered rework is central to the management of complex projects.
Open source video
Project dynamics emerge endogenously from interactions between workforce, schedule, tasks, and perceived progress.
Open source video
halting hiring when projects plateau prematurely
Open source video
Source video

Evaluation horizons sized to the slowest compensating feedback loop

Research question

How can the time constant of a compensating feedback loop be estimated empirically before knowing it, so that evaluation windows for an intervention are set long enough that benefit rankings do not invert?

Why

Short-horizon wins are systematically selected for when compensating loops exist: added capacity, raised timeouts, relaxed thresholds, or added agents all show immediate improvement while the compensating loop erases it later. Without a method for bounding the loop timescale, benchmark duration is chosen by convenience and the ranking of interventions is wrong.

Policies like building low-cost housing or jobs programs can have unintended long-term compensating feedback effects.
Open source video
a successful jobs program temporarily helps a city, but long-term feedback loops negate the effect
Open source video
Source video

Noise-excited oscillation in delayed control loops (autoscaling, retries, capacity adjustment)

Research question

Do delayed software control loops exhibit a spectral peak at a natural frequency under white-noise input, and does input smoothing leave that peak intact while retuning the loop removes it?

Why

If oscillation frequency is endogenous to loop delay, then input damping (rate limits, jitter, traffic smoothing) cannot fix flapping and retry storms and can mask the loop that actually needs retuning. This converts a common operational misdiagnosis into a testable spectral claim.

Random disturbances (noise) can cause a system to oscillate at its natural frequency.
Open source video
Simple system dynamics models with inventory, production capacity, and delays can reproduce oscillatory behavior.
Open source video
Source video

Minimal causal-structure models as an intervention-finding tool versus black-box fidelity models

Research question

For a system whose goal is intervention rather than prediction, does a minimal stock-and-flow abstraction (inventory, capacity, delay) locate the effective intervention point better than a higher-fidelity fitted model?

Why

The abstraction is what makes the loop legible and the intervention point identifiable. If minimal structures reproduce the observed behavior, then model complexity is not the gate on decision usefulness — legibility and modifiability are.

Simple system dynamics models with inventory, production capacity, and delays can reproduce oscillatory behavior.
Open source video
Visualizing inventory and commodity flows using the mental model of a bathtub with inflow pipes and outflow pipes.
Open source video
Source video

Article ideas

4

Your Progress Metric Is Lying: Undiscovered Rework in Agentic Pipelines

Control loops that allocate effort using self-reported or partially verified progress will systematically accumulate hidden rework, and the correct fix is not better estimation but instrumenting perceived and verified progress as two separate stocks with an explicit rework flow back into the work queue.

Angle

Map project-dynamics rework accounting onto agentic/software delivery: show how 'task marked done', 'tests passing', and LLM self-reported completion act as perceived-progress proxies, why capacity decisions keyed to them plateau prematurely, and what a dual-signal harness looks like.

Source video

Benchmarks Must Outlive the Feedback Loop

Any intervention evaluated on a horizon shorter than the slowest compensating feedback loop it triggers will be ranked by its transient, not its steady state, so evaluation horizon is an architectural parameter that must be derived from the system being measured rather than chosen for convenience.

Angle

Argue that short-horizon benchmark wins (added capacity, raised timeouts, more agents, relaxed thresholds) invert in ranking once the compensating loop acts; propose estimating loop time constants by perturbation and benefit-decay curves.

Source video

Stop Smoothing Your Traffic: Noise Doesn't Cause Your Oscillations

Oscillatory instability in delayed control loops is excited by noise but produced by structure, so input smoothing dampens symptoms while leaving the natural frequency intact — the correct remediation is identifying and retuning the delay-bearing loop.

Angle

Operational debugging narrative: latency spikes, retry storms, and autoscaler flapping are routinely blamed on traffic; show the spectral-peak diagnostic and contrast smoothing versus loop retuning as interventions.

Source video

Explain Failure From Your Own Structure, Not Your Users

Endogenous explanation — deriving behavior from internal loops, contracts, and policy rather than external shocks or model variance — is the discipline that makes reliability work actionable, because external-cause explanations are unfalsifiable and therefore unactionable.

Angle

Apply Forrester-style endogenous modeling to AI system reliability: attribute failure to retry policies, tool contracts, and context assembly instead of to input variance, and show what changes in what you fix and what you measure.

Source video

Project ideas

3

Dual-Signal Rework Detector

gatehouse

In a long-running agent task pipeline, the gap between perceived completion (agent-declared done) and independently verified completion at time T predicts late-stage rework volume at T+n, and gating capacity decisions on verified progress reduces total rework without reducing throughput.

Proof of concept

Instrument a multi-step agent workflow with two parallel progress stocks: a self-reported completion counter and an independent verification pass. Log the divergence series, and compare a control policy that scales agents on perceived progress against one that scales on verified progress.

Measurement

Divergence rate between perceived and verified completion per interval versus rework volume in later intervals; total rework and wall-clock time under each scaling policy.

Source video

Compensating-Loop Horizon Estimator

beyond-evals

For a given intervention (adding capacity, raising a timeout, relaxing a threshold), the measured benefit decays over time with a time constant matching the slowest compensating loop, and evaluations shorter than that constant invert the ranking of interventions.

Proof of concept

Perturb one capacity/retry parameter in a running system, then track the benefit decay curve against a control arm; fit an exponential decay to estimate the compensating loop time constant, then re-rank a set of candidate interventions using horizons above and below that constant.

Measurement

Fitted decay time constant of intervention benefit; rank correlation of intervention ordering between short-horizon and long-horizon evaluations.

Source video

Control-Loop Spectral Diagnostic

movement-lab

A delayed capacity-control loop (autoscaler, retry controller, scheduler) exhibits a spectral peak at its natural frequency under white-noise input, and input smoothing leaves that peak intact while retuning loop gain/delay removes it.

Proof of concept

Stand up a controllable delayed capacity loop with tunable gain and delay; inject white noise at the input; run two arms — input smoothing versus loop retuning — and compare power spectra and oscillation amplitude of the controlled variable.

Measurement

Power spectral density peak location and magnitude at the natural frequency across arms; time-domain oscillation amplitude before and after each intervention.

Source video

Architectural implications

3

Progress signals that are self-reported or partially verified diverge from real completion, and control loops act on the proxy.

Before

Effort/progress tracked as a single 'done' metric; capacity and parallelism decisions made from it.

After

Track perceived progress and verified progress as separate stocks with an explicit rework flow back into the work queue.

Consequence

Reveals hidden rework accumulation before it becomes a late-stage schedule or cost explosion; prevents premature plateau-driven decisions.

Source video

Compensating feedback loops reverse the sign of intervention benefit over long horizons.

Before

Evaluation windows sized by convenience (short runs, single-turn benchmarks).

After

Evaluation horizon set by the slowest known compensating loop in the architecture.

Consequence

Interventions ranked differently, and short-horizon 'wins' no longer ship as architectural changes.

Source video

Oscillation frequency is endogenous and excited by noise.

Before

Damp inputs (rate limits, smoothing, jitter) to suppress instability.

After

Identify the delay-bearing control loop and retune it; treat noise as excitation, not cause.

Consequence

Fixes flapping/retry storms at the structural level rather than masking them at the boundary.

Source video

Tradeoffs and failure modes

3

Perceived vs real progress in project control

Benefit

Perceived-progress signals are cheap, fast, and actionable, enabling continuous management decisions.

Cost or risk

They diverge from real progress and hide undiscovered rework, causing premature plateauing and eventually large overruns.

undiscovered rework is central to the management of complex projects
Open source video
Source video

Single-point policy intervention in dense feedback systems

Benefit

Produces immediate, measurable improvement (jobs program helps the city; added capacity lowers latency).

Cost or risk

Long-term compensating feedback negates the effect, so the intervention is adopted on the strength of a transient.

a successful jobs program temporarily helps a city, but long-term feedback loops negate the effect
Open source video
Source video

Minimal models vs high-fidelity models

Benefit

Minimal stock-and-flow structures remain interpretable and reproduce observed oscillation.

Cost or risk

They necessarily omit structure; the summary does not establish how far minimal models transfer to new domains.

Simple system dynamics models with inventory, production capacity, and delays can reproduce oscillatory behavior.
Open source video
Source video

Open questions

4

Can 'undiscovered rework' be detected online in software/agentic pipelines, or only diagnosed retrospectively as cost/schedule variance?

Why unresolved

The summary asserts rework is central and typically undiscovered but provides no detection mechanism or instrumentation.

Research direction

Build a dual-signal harness (perceived vs independently verified completion) on a long-running agent task and measure divergence rate vs later rework volume.

Source video

What is the minimum evaluation horizon needed to expose compensating feedback in an engineered system, and how should it be set without already knowing the loop timescale?

Why unresolved

The summary demonstrates the phenomenon qualitatively but gives no method for bounding the slow loop's timescale.

Research direction

Perturb a capacity/retry parameter and track benefit decay curves to estimate the compensating loop time constant empirically.

Source video

Which of the 'impact' factors (problem definition, stakeholder involvement, multiple perspectives, requisite modeling, sensitive communication) actually determines whether model output changes a decision?

Why unresolved

The summary lists them as a set without weighting, ablation, or evidence separating their contributions.

Research direction

Compare deployments that differ on one factor (e.g., decision-risk communication layer present vs absent) against decision-change outcomes.

Source video

Does noise-excited natural-frequency oscillation appear in autoscaling and retry control loops, and does input smoothing reliably fail to suppress it?

Why unresolved

The claim is supported only for commodity markets in the summary; transfer to software control loops is inferred, not demonstrated.

Research direction

Instrument a production autoscaler for spectral peaks and test whether smoothing vs loop retuning removes the oscillation.

Source video

Key claims

8
causalVerification needed

Random disturbances can drive a system to oscillate at its natural frequency.

Evidence

Random disturbances (noise) can cause a system to oscillate at its natural frequency.

Question

In a delayed capacity-control loop, does white-noise input produce a spectral peak at the loop's natural frequency, and does input smoothing leave that peak intact?

Source video
causalVerification needed

Undiscovered rework is central to managing complex projects.

Evidence

Undiscovered rework is central to the management of complex projects.

Question

What fraction of late-stage project cost overrun is attributable to rework that was not visible in perceived-progress reporting at the time?

Source video
causalVerification needed

Project dynamics arise endogenously from interactions among workforce, schedule, tasks, and perceived progress.

Evidence

Project dynamics emerge endogenously from interactions between workforce, schedule, tasks, and perceived progress.

Question

Can overrun trajectories be reproduced from endogenous workforce/schedule/progress interactions alone, without exogenous shocks?

Source video
causalVerification needed

Well-intentioned policies such as jobs programs or low-cost housing can be undone by long-term compensating feedback.

Evidence

Policies like building low-cost housing or jobs programs can have unintended long-term compensating feedback effects.

Question

Over what time horizon does the measured benefit of such an intervention invert, and does that match the modeled compensating loop's time constant?

Source video
causalVerification needed

System dynamics models were decisive in moving WHO from polio outbreak containment to a global eradication strategy.

Evidence

System dynamics models played a crucial role in convincing the WHO to shift from outbreak containment to global eradication of polio.

Question

What specific decision documents or policy shifts can be attributed to the model output rather than to other advocacy?

Source video
factualVerification needed

System dynamics models were used in litigation to adjudicate large cost overruns (Litton/US Navy; Channel tunnel).

Evidence

Litton industry ship building litigation with the US Navy

Question

In which rulings or settlements were system dynamics model outputs admitted or relied upon as evidence?

Source video
causalVerification needed

Simple stock-and-flow models with inventory, production capacity, and delays reproduce observed commodity price oscillation.

Evidence

Simple system dynamics models with inventory, production capacity, and delays can reproduce oscillatory behavior.

Question

Does a three-stock model with a single capacity-adjustment delay reproduce the amplitude and period of observed commodity price cycles out of sample?

Source video
opinionVerification not requested

Impactful modeling depends on problem definition, stakeholder involvement, multiple perspectives, requisite modeling, and sensitive communication.

Evidence

emphasizing the importance of significant problem definitions, clear dynamic problem formulation, stakeholder involvement, multiple perspectives, requisite modeling, and effective sensitive communication

Source video

Connections

5