Skip to content
The People's
Register
Reporting. Analysis. Public information.
Menu

Ideas & History · Biology, models and public decisions

Richard Levins: a model is an argument about what matters

Richard Levins used deliberately simple models to expose failures of common sense. His work also asks a harder question: who decides what a successful intervention should achieve?

A pest-control programme can look successful at the boundary of a field and still fail at the boundary of a landscape. That difference is more than a question of scale. It changes what must be explained, which observations matter and what counts as success. Richard Levins made this problem productive: a model should help researchers discover which relationships they have overlooked before its tidy answer becomes an instruction to act.

The landscape outside the calculation

In his 1969 paper on environmental heterogeneity and biological control, Levins modelled a region containing distinct local pest populations. Local extinction could coexist with regional persistence because insects recolonised empty sites. He distinguished interventions affecting extinction from those affecting migration. Under his assumptions, the two approaches could favour opposite spatial patterns of control. Before applying the model, researchers needed evidence of discrete populations, distance-dependent migration and a delay before reinvasion. [2]

The achievement is a change in the question. A field inspection asks whether an intervention reduced the population here. A regional assessment asks whether the intervention altered the conditions under which infestation returns. Those questions can both be legitimate, but an answer to the first cannot be advertised as an answer to the second. An institution that pays only for the most immediately visible result has already encouraged a particular account of success. A longer view may expose work that looks unproductive in a short reporting period but is essential to sustaining the result.

Levins also separated possible objectives: eliminating a pest, reducing its average abundance, limiting crop damage, or balancing control costs against damage. Biology alone could not choose among them. His conclusion was explicitly limited: simplified models could reveal possibilities and challenge apparently obvious assumptions, but they could not prescribe a control strategy or replace knowledge of the insects concerned. [2]

This distinction prevents mathematical competence from quietly becoming political authority. The best intervention for a commercial grower’s balance sheet need not be the best one for workers, neighbouring farms or public expenditure. That is not an allegation about any particular pesticide or farming system. It follows from defining different objectives. If a decision promises a public benefit, the people evaluating it need to know whose gains and losses have entered the objective and whose have been left outside. A precise answer to a privately chosen question does not, by precision alone, acquire a public mandate.

Simplification and its critics

Richard Levins made an uncomfortable argument in 1966: a biological model could not simply maximise generality, realism and precision together. Researchers had to make choices. A model useful across many settings might sacrifice detail; a detailed account of one setting might travel badly. His paper, The Strategy of Model Building in Population Biology, treated those choices as part of scientific judgement. [3]

That influential formulation has faced a serious challenge. In 1993, Steven Hecht Orzack and Elliott Sober argued that the three terms needed clearer definitions and that there was no necessary conflict among them. They challenged the claim as a necessary feature of modelling, rather than merely a recurrent practical difficulty. [4]

Levins’s reply distinguished model building as an evolving research practice from the classification of finished mathematical objects. His three strategies described tendencies in population biology, not every possible model. Crucially, he distinguished maximising all three qualities from improving them together. He explicitly accepted that a change could improve all three. He also rejected a definition of realism based simply on adding independent variables. A model could change by introducing interactions, relaxing an assumption or restricting the circumstances to which it applied. What made a simplification useful depended on the scientific question. [5]

There is something to preserve on each side of this dispute. Research has finite time, evidence and resources, so choices about representation cannot be wished away. The dispute cannot be settled simply by showing that a weak model can be improved in several respects. A public modeller should be able to explain an actual compromise: which feature was omitted, why retaining it was impractical, and whether that omission could reverse the recommendation. Invoking complexity in the abstract cannot substitute for this account. Nor should criticism demand an exhaustive model that would make any decision impossible.

The distinction between understanding and forecasting matters here. An explanation may identify a mechanism without telling a farmer exactly how large next season’s loss will be. Conversely, a useful forecast may predict an outcome without identifying the intervention that would change it. Neither achievement should be dismissed. The danger arises when the institution purchasing one quietly presents it as the other. An intelligible model can improve debate even when it cannot produce a reliable number; a reliable number still needs an argument connecting it to the proposed action.

Agreement is not an experiment

Orzack returned to another disputed claim in 2011: the status of conclusions shared by several models. He questioned whether models of the same phenomenon could be treated as independent evidence. He also distinguished agreement across models from a prediction’s insensitivity to changing parameters within one model. Neither property, he argued, automatically established that the prediction was true of nature. [6]

This objection is particularly valuable when several technical reports support the same policy. Their agreement may be informative, but it matters whether they rely on the same doubtful premise. Commissioning three analyses does not create three independent observations. Equally, disagreement can be useful if it reveals the assumption on which the decision turns. The appropriate response is to locate that assumption and seek evidence capable of challenging it. Public discussion should distinguish confidence produced by repeated calculation from confidence produced by confronting a prediction with the world.

A further distinction concerns when evidence is collected. Checking a model against information already used to construct it can expose errors, but it is a different test from confronting it with observations held back from that construction. An institution must also decide what to do when results are mixed. Revising a model can be intellectually responsible; repeatedly changing the definition of success to protect a favoured policy is not. Accountability requires a record of the original claim and the evidence that prompted its revision. Otherwise technical flexibility becomes a way to evade political responsibility.

Who chooses the problem?

Levins’s 2008 autobiographical essay places these methodological concerns within a political life. He contrasted learning about agricultural surplus at Cornell with farming in a poor region of Puerto Rico, where agriculture’s significance for people’s lives became immediate. He described collaboration with Rosario Morales on an agrarian programme, and later work in Cuba. The essay presents scientific and socialist commitments as connected. It also acknowledges a profound early error: his youthful interest in Lysenko did not justify deriving biological conclusions from philosophical principles. [7]

That admission matters to a radical reading of science. A commitment to liberation can direct attention towards neglected suffering and unequal control of resources. It cannot establish which biological mechanism operates, any more than corporate sponsorship can make a favourable result true. Political commitment becomes a source of intellectual strength when it makes the choice of problems explicit and accepts the risk that evidence will defeat a preferred explanation. Treating an empirical disagreement as disloyalty destroys precisely the capacity to learn that collective action needs.

For South African public institutions, the resulting demand is concrete. A model used to justify an agricultural programme should expose the objective, the boundary of the calculation and the conditions under which the recommendation would change. People affected by the decision need the means to dispute those choices, including access to the evidence and an explanation they can interrogate. Their participation does not replace specialised knowledge. It makes the purpose and consequences of deploying that knowledge answerable to those who must live with them.

Levins is most useful when his work makes such disagreement more exact. The field, the landscape, the balance sheet and the livelihood are connected, but they are not interchangeable objects of inquiry. Choosing among them is part of building the model. Making that choice visible allows a technical argument to become a democratic one without asking either mathematics or political conviction to do work it cannot perform.

About the desk

Analysis and public scholarship

The News Desk reports and analyses public affairs, science, culture and history for The People’s Register.

Profile and published work →
Sources and further reading
  1. Harvard, Richard Levins memorial (opens in a new tab) Published 27 January 2016.
  2. Richard Levins, Some Demographic and Genetic Consequences of Environmental Heterogeneity for Biological Control (opens in a new tab) Bulletin of the Entomological Society of America 15 (1969), pp. 237–240.
  3. Richard Levins, The Strategy of Model Building in Population Biology (opens in a new tab) American Scientist 54 (1966), pp. 421–431.
  4. Steven Hecht Orzack and Elliott Sober, A Critical Assessment of Levins’s The Strategy of Model Building in Population Biology (opens in a new tab) Quarterly Review of Biology 68 (1993), pp. 533–546.
  5. Richard Levins, A Response to Orzack and Sober: Formal Analysis and the Fluidity of Science (opens in a new tab) Quarterly Review of Biology 68 (1993), pp. 547–555.
  6. Steven Hecht Orzack, The philosophy of modelling or does the philosophy of biology have any use? (opens in a new tab) Philosophical Transactions of the Royal Society B (2011).
  7. Richard Levins, Living the 11th Thesis (opens in a new tab) Monthly Review 59, no. 8 (January 2008). Autobiographical testimony.

Request a correction · Editorial standards