Oskar Morgenstern’s scholarly essay, published as an individually titled contribution to an edited volume, proposes a reorganization of economic inquiry around controlled experiment and electronic computation. Its seven sections move from the history of numerical economic theory through the relationship between computation, experiment, and measurement to concrete research programs. The culminating proposal is that computers might generate unfamiliar empirical material, extending economics beyond both unaided observation and the numerical application of existing theory. Computational speed matters because sufficiently large increases change what economists can investigate, not merely how quickly they obtain familiar results.
Morgenstern begins by locating these possibilities within a crisis of static equilibrium theory. Game theory and linear programming have already challenged the treatment of economic conduct as ordinary maximization; experiment and computation promise a further transformation. Mathematical sophistication alone, however, cannot secure substantive validity. Economics must adopt the empirical discipline of advanced sciences without assuming that its models should reproduce their theoretical structures.
His historical account distinguishes mathematical formulation, proof of meaningful solutions, and numerical execution. Walras’s achievement in representing interdependence remained empirically weak, while counting equations and unknowns could not establish solvability. Wald and von Neumann introduced more demanding standards of existence proof; Leontief’s input-output calculations subsequently demonstrated numerical application. These advances establish a new expectation:
It may well be said that a modern economic theory will now no longer be considered finished until it is clearly shown how it can give numerical results.
Numerical applicability nevertheless requires more than computing machinery. Observational errors, rounding, and the mathematical properties of particular systems can invalidate results even where an abstract solution exists. Morgenstern also exposes a revealing asymmetry: equilibrium theorists attribute extraordinary information and calculating capacities to individuals while treating numerical application of their own systems as impracticable. Nor does saying that reality “solves” the equations establish that a particular model accurately represents it.
The essay’s conceptual hinge is a distinction between experiments seeking general properties through theory-guided manipulation and experiments primarily discovering individual facts. For the first class, Morgenstern advances a strong equivalence:
We can now state a general thesis: Every computation is equivalent to an experiment of the first kind and vice versa.
An experimental arrangement can function as an analogue computer, physically representing a problem whose explicit mathematical solution may be unavailable. Conversely, sufficiently developed theory and numerical methods can replace costly physical trials. The equivalence is conditional on adequate representation and accuracy: a hydraulic economic model is not a reliable research instrument merely because it vividly illustrates interdependence. Digital computers offer broader applicability, but their superiority does not eliminate the need for experiment. Scientific advances continually disclose further questions requiring observation.
Morgenstern then unsettles the conventional opposition between observational and experimental sciences. Measurement itself involves controlled physical operations, instruments, uncertainty, and decisions about accuracy. Astronomy consequently supplies a methodological parallel for economics despite its inability to manipulate stars. Commercial routines and quality control can likewise possess experimental properties irrespective of their immediate motives. This argument widens the field of usable evidence without licensing the treatment of every policy intervention as a well-designed scientific experiment.
The extensive survey of experimental possibilities distinguishes direct interventions from indirect, computational experiments. Thünen exemplifies sustained cooperation between observation and theory; business trials of prices, packaging, advertising, wages, and working arrangements offer largely neglected opportunities. Government measures and historical communities are harder to interpret because initial conditions and outcomes were often inadequately recorded. Morgenstern entertains deliberately organized experimental communities while acknowledging ethical restrictions, observational disturbance, and interpretive difficulty. More narrowly controlled investigations—especially market and utility experiments associated with Chamberlin, Mosteller and Nogee, and Edwards—may advance theory more securely than aggregate interventions.
Indirect experiments use observed or hypothetical parameters to explore models and compare computed outcomes with observations, including retrospective forecasts. Morgenstern surveys military planning, transport, inventories, assignment problems, and optimal strategies before emphasizing scientific applications. Repeated computations can vary parameters, compare models, and investigate aggregation where theory remains incomplete. Better business computation also changes the economy being studied: improved information and coordination affect actual conduct rather than simply describing it.
The most speculative section proposes computation without prior guidance from a developed economic theory. Like telescopes and microscopes, computers might disclose relations inaccessible to ordinary intuition. Fourier analysis, autocorrelation, and large-scale correlations could produce new material for subsequent hypothesis formation. Such inquiry still depends on mathematics, statistics, and instrument design; its independence concerns economic explanation, not theory altogether. Morgenstern explicitly limits the promise:
From the lack of sufficient guidance by economic theory it follows that not all the numbers obtained are "data," i.e., that any meaning attaches itself to them.
Computational output therefore requires interpretation, significance testing, and potentially new methods of analysis. The conclusion returns to elementary observations as the indispensable foundation. Experimental data and computationally generated relations should develop together, supported by teams, institutions, funding, and closer contact with data-collecting agencies. The essay’s enduring relevance lies in joining computational ambition to empirical restraint: machines can expand the objects and methods of economics, but cannot themselves establish what their results mean.
This work was divided into 13 sections when it entered the library's research corpus—an apparatus for search and citation, not necessarily the author's own table of contents. Each title opens its summary.
Put a question to this work; the Librarian answers from its 13 sections and cites the passage.
Ask the Librarian