A New Framework Tests Circular-Economy Policies Against Unknowns

Policies designed to keep materials in circulation often depend on models that appear more precise than the evidence allows. A new framework for material flow analysis seeks to address that problem by asking a different question: not which policy is most likely to succeed, but which policies remain reasonably effective when the future cannot be described with trustworthy probabilities. The approach, developed by Norbert Weijenberg, Willem Auping, Sietske Lensen, Anna Schwarz and Nils Thonemann, combines material flow analysis with exploratory modeling and analysis. The researchers describe the method as a way to evaluate policy robustness under “deep uncertainty,” a situation in which experts lack sufficient evidence to agree on probability distributions, model structures, system boundaries or even the outcomes that matter most. Their illustrative application examined plastic pollution from polyethylene terephthalate, or PET, beverage bottles in the Netherlands. The authors stress that the case study was designed to demonstrate the method rather than provide policy recommendations. Even so, it shows how decision-makers could compare interventions while making uncertainty visible instead of hiding it behind apparently exact averages.

Material flow analysis tracks how substances and products move through a defined system, from production and consumption to recycling, disposal and environmental release. It has become an important tool for studying circular-economy strategies, material efficiency and pollution pathways. Yet its results are only as reliable as the data and assumptions built into the model. In data-rich settings, analysts may estimate probability distributions from repeated measurements or large datasets. In many real-world material systems, however, data are sparse, noisy or assembled from proxies, expert judgments and isolated observations. Assigning a probability distribution in those circumstances can imply knowledge that researchers do not actually possess. Conventional methods may also produce symmetric uncertainty ranges for variables whose real behavior is strongly skewed, or generate mathematically impossible values for transfer coefficients that must lie between zero and one. The new framework therefore replaces the search for the most probable outcome with a broad stress test. It explores plausible combinations of conditions, including unfavorable ones, and identifies policies that perform acceptably across that range.

The framework organizes a material-flow model using the XLRM structure: external factors, levers, relationships and performance metrics. External factors include uncertain quantities such as material imports, flow magnitudes, product lifetimes and transfer coefficients, which describe the fraction of material moving from one process to another. Levers represent changes that a policy can influence, while relationships capture uncertainty about model structure or normative choices, such as which material categories should be included or which processes belong inside the system boundary. Performance metrics describe what success means, including accumulated stocks, material flows or recycling rates. The researchers then evaluate each policy under many combinations of these elements. A policy is considered robust when it performs reasonably well relative to alternatives across a wide range of plausible system behaviors and value perspectives. Rather than relying on one summary statistic, the method can use several robustness measures. Satisficing measures count how often a policy reaches a minimum threshold, regret measures compare it with the best alternative in the same scenario, and statistical measures summarize the distribution of outcomes.

A central technical challenge concerns transfer coefficients. At every process in a material-flow model, the fractions leaving the process must add up to one, preserving mass balance. Sampling each coefficient independently can violate that rule, while normalizing independently sampled values afterward may distort the uncertainty that experts intended to express. The researchers address this problem in two stages. First, a constrained elicitation procedure asks experts to define lower and upper plausible bounds while respecting the sum-to-one relationship from the beginning. Second, a multivariate sampling method based on truncated, flat Dirichlet distributions generates values inside those bounds without requiring post hoc normalization. A two-blocked Gibbs sampling procedure is used to approximate the constrained distributions efficiently. For vulnerability analysis, the framework applies Spearman’s rank correlation coefficient to measure how strongly uncertain factors are associated with policy performance. This method is comparatively simple and interpretable, although the authors acknowledge that it may miss higher-order interactions. Such interactions can arise across an entire material-flow network because transfer coefficients are multiplied along chains of processes.

To illustrate the approach, the team examined PET consumer beverage bottles in the Netherlands from 2025 through 2030. The model followed annual flows from consumption through collection, recycling, incineration, landfilling and environmental release, distinguishing macroplastics larger than five millimeters from microplastics smaller than five millimeters. The analysis focused on two objectives: reducing cumulative environmental microplastic emissions and increasing cumulative secondary material production from used PET bottles. The researchers evaluated eight policy proposals alongside a business-as-usual case. These included measures represented in the model as changes in collection, consumption or recovery flows, such as return incentives, consumption reduction, recovery at recycling plants and extraction from landfills. Because policy effects were not elicited from experts, each affected factor was adjusted by a uniform 10 percent relative to business as usual. The authors explicitly state that this assumption was a simplified device for demonstrating model behavior, not a realistic estimate of policy effectiveness. The analysis used 136 uncertain external factors, 500 sampled scenarios and nine policy alternatives, producing 4,500 policy-scenario experiments.

The simulated outcomes showed why robust evaluation can look different from conventional policy ranking. Across the scenarios, cumulative microplastic emissions ranged approximately from 10 to 130 tonnes, while secondary material production ranged from about 50 to 550 tonnes. The broad ranges reflected the deliberately wide uncertainty bounds rather than a forecast of expected national outcomes. The return policy achieved the best best-case result for microplastic emissions and was consistently the most robust option for secondary material production across the selected metrics. The reduce policy produced the best worst-case outcome for microplastic emissions, suggesting a more risk-averse profile, but it also had a higher failure rate for that objective and performed worst for secondary material production. The trade-off follows directly from the material balance: reducing PET consumption can limit pollution, but it also reduces the quantity of material available for recycling. A capture policy aimed at recovery in recycling plants showed moderate robustness for both objectives, while an extract policy targeting landfill recovery was moderately robust for emissions but among the least robust for secondary material production because relatively little PET flowed through landfills.

The vulnerability analysis identified the factors most closely associated with policy failure or success. The uncertain share of PET bottles entering deposit-return systems strongly influenced performance. The fractions of plastic recycled, incinerated or lost during packaging recycling were especially important for secondary material production. Microplastic-emission outcomes were additionally sensitive to where bottles were consumed, emissions associated with on-the-go use, dispersion from indoor air, the application of contaminated compost to agricultural soil, uncollected mixed waste reaching surface waters and direct incineration of collected packaging. These variables are “robustness controls” in the framework because improving knowledge about them could make policy comparisons more reliable. They are not automatically leverage points for intervention. A factor may appear influential only because its uncertainty range is wide, while a currently minor factor could become important if a policy changes the system outside the tested range. In two repeated runs, robustness scores showed good to excellent agreement after min–max normalization, but absolute scores were unstable with only 500 scenarios. That result underlines the need for convergence testing and larger samples when computational resources permit.

The researchers say the framework is best suited to focused material-flow models in which probabilities are genuinely difficult to justify. It is not a replacement for probabilistic analysis when extensive empirical data support credible distributions, or when decision-makers specifically need the most likely outcome. Its practical limitations are substantial: expert elicitation becomes burdensome as the number of processes grows, and large models can make broad scenario exploration computationally expensive. The demonstration relied on one methodologically knowledgeable expert, left some model relationships and material imports fixed, and did not fully represent uncertainty in policy effects or indirect consequences. The authors propose reducing the burden by beginning with broad ranges and concentrating expert attention on the most important robustness controls. Future work could incorporate uncertainty in policy levers, evaluate combinations of interventions, improve sampling efficiency and use statistical techniques capable of detecting higher-order interactions. By shifting attention from false precision to transparent stress testing, the framework offers a way for circular-economy policy analysis to remain useful even when the evidence is incomplete and the future refuses to behave like a probability distribution.

The framework also clarifies why uncertainty in a material-flow model is not a single quantity. Numerical uncertainty concerns values such as imports, lifetimes or transfer coefficients, whereas structural uncertainty concerns how the system itself is represented. Analysts may disagree about whether a process belongs in the model, which material categories are relevant, or how a flow should be interpreted. These choices can alter policy rankings even when the numerical inputs remain unchanged. Treating alternative structures and value perspectives as explicit scenarios therefore makes disagreements inspectable rather than burying them in one composite estimate.

Its robustness perspective changes what counts as useful evidence. A factor associated with policy failure is not necessarily the best target for intervention; it may simply have been assigned a particularly broad plausible range. Conversely, a factor with a modest apparent influence could become decisive under a different policy or model structure. Robustness controls are therefore most useful for prioritizing data collection, expert review and model refinement. They can also reveal when a conclusion depends on a narrow set of assumptions, which is important before translating a model result into a regulatory claim.

The approach is consequently complementary to, rather than universally superior to, probabilistic uncertainty analysis. Where repeated observations justify defensible likelihoods, probability-based results can answer questions about expected outcomes and risk frequencies. Under deep uncertainty, however, the framework avoids presenting unsupported probabilities as measured facts. Its value lies in preserving the material-balance logic of MFA while widening the analysis to include plausible alternatives, adverse conditions and competing definitions of success. That combination can help decision-makers identify strategies that are not optimal in every modeled future, but are less vulnerable to being undermined by an incorrect assumption.

Subject of Research: Robust policy evaluation in material flow analysis under deep uncertainty

Article Title: Robust policy evaluation in material flow analysis under deep uncertainty

Article References: Weijenberg, N., Auping, W., Lensen, S., Schwarz, A., & Thonemann, N. (2026). Robust policy evaluation in material flow analysis under deep uncertainty. Journal of Industrial Ecology. https://doi.org/10.1007/s44498-026-00170-5

Image Credits: AI Generated

DOI: 10.1007/s44498-026-00170-5

Keywords: material flow analysis, deep uncertainty, exploratory modeling, circular economy, plastic pollution, PET recycling, policy robustness, uncertainty analysis, Robust, policy, evaluation, material