Oskar Morgenstern, “Discussion” (1949)

Oskar Morgenstern’s published conference discussion contribution, spanning pages 238–240, examines the conditions under which input-output analysis can produce meaningful economic results. Responding principally to Professor Leontief, he welcomes the promise of electronic computation while challenging the assumption that greater calculating power can overcome unreliable observations, unstable solutions, or inadequate models. His central argument is that solving a system of equations and obtaining useful economic knowledge are distinct accomplishments. The contribution moves from a classification of mathematical and computational difficulties to the quality of economic data, then considers particular features of input-output tables and production assumptions before concluding with a call for methodological care.

Morgenstern begins by placing economic computation at the frontier of contemporary numerical practice. Large systems of simultaneous equations confront economists with difficulties for which, he argues, even the natural sciences offer little precedent. Electronic computers promise operations previously beyond practical reach, but their arrival does not remove the need to understand the conditions governing a solution’s reliability. Problems arise even with ten equations and entirely elementary operations; dynamic models involving differential equations would introduce further complications. His references to investigations by von Neumann, Goldstine, and others make numerical analysis integral to the economic undertaking rather than a technical service supplied after the substantive work is finished.

He distinguishes three difficulties: whether the problem has been given an appropriate mathematical formulation, whether its parameters are accurately known, and whether the computational process itself introduces obstacles. Separately, he identifies the question of whether the underlying, essentially Walrasian economic model adequately describes reality. A mathematically satisfactory formulation need not be an empirically satisfactory model, while an intuitively persuasive model may resist available mathematical treatment. Morgenstern brackets this broader controversy to concentrate on a more immediate requirement:

The main problem, therefore, is for us to find out whether the data are good enough for these purposes.

This question is decisive because the stability of the mathematical solution, and therefore the usefulness of the enterprise, depends on the quality of its inputs. Morgenstern criticizes an economic culture that treats faithful transcription or elaborate numerical presentation as evidence of accuracy. Neither establishes how closely an observation corresponds to the quantity being measured.

Often accuracy is identified with detail; for example, the giving of figures with many decimal points when the real inaccuracy has already been permitted to enter into a much rougher figure.

The contrast is between displayed precision and warranted knowledge. Morgenstern reports that Princeton investigations of routinely used economic information produced “absolutely shocking” results, although he supplies no detailed findings here and defers their presentation to future publications. His positive requirement is nevertheless clear: economists must collect better data and establish observational errors as fully as possible. Only then can matrix inversion and the stability of its results be assessed rigorously. The passage thus proposes a research priority without offering evidence sufficient to evaluate the reported investigations independently.

The discussion next turns to the structure of existing input-output tables. Their numerous zero entries suggest that industries transact with relatively few others, but Morgenstern asks whether this pattern reflects defective information or industrial classification. He also notes the substantial undistributed amounts mentioned by Leontief. These features matter because they can make inversion easier and the resulting solution more stable than a more fully populated matrix would be. Apparent computational success therefore cannot, by itself, validate the table: the properties facilitating calculation may partly arise from deficiencies in the observations.

His treatment of constant production coefficients is more accommodating. He agrees with Leontief that this assumption may be less damaging than critics suppose, notes Koopmans’s methods for addressing increasing and decreasing returns, and argues that moving immediately to nonlinear relationships would compound already immense difficulties. Yet this is a provisional methodological judgment, not a claim that constancy has been established.

Fundamentally, this is again an observational-empirical issue and the answer hinges once more upon the accuracy of our observations.

Neither acceptance nor rejection of constant coefficients can rest securely on general impressions. The same insistence on evidential discipline informs his brief challenge to Leontief’s description of customary supply and demand curves as noncontroversial. Morgenstern invokes unresolved problems of non-additive curves and questions the universality of textbook constructions. He leaves these model-side issues largely undeveloped because securing good data with known observational errors comes first.

The conclusion combines support for input-output research with a warning against automatic confidence in its outputs. Morgenstern explicitly includes himself among economists proposing such undertakings, but insists that they learn the logic of computing rather than simply entrust equations to machines.

It would not be advisable to raise hopes that perhaps cannot be fulfilled.

The contribution’s relevance lies in its separation of computational capacity from epistemic warrant. Better machines can make an ambitious calculation feasible without establishing that its result is stable, empirically grounded, or economically meaningful. Morgenstern’s desired combination of boldness and care therefore entails attention to observations, mathematical formulation, and numerical procedure together. His final emphasis is unambiguous:

I raise my voice in favor of great care.