Oskar Morgenstern, “Qui Numerare Incipit Errare Incipit” (1963)

Oskar Morgenstern’s magazine article, presented as his condensation of the forthcoming revised On the Accuracy of Economic Observations, argues that economic statistics routinely claim more precision than their evidence warrants. Moving from general problems of observation to unemployment, national income, international comparisons, and growth rates, it culminates in a demand for published error estimates and theories adapted to imperfect information. Its target is not quantification itself but the conversion of uncertain measurements into apparently exact grounds for business decisions and government policy.

The opening comparison with physics establishes the governing standard: economic observations cannot escape the errors encountered in natural science, yet economists devote far less attention to them. In physics, precision is constrained by measurement procedures and theoretical requirements; public economic reporting instead treats additional digits as additional knowledge. Price indexes, unemployment totals, and estimates of spending power circulate at levels of refinement that their underlying observations cannot sustain. Such precision matters politically because wage agreements and judgments about inflation or government performance depend upon it.

The public in general and Congress in particular must be made to understand that there cannot be absolute accuracy, that there must be error, and that the important thing to do is to try to uncover, remove, or at least limit the error.

Morgenstern distinguishes unavoidable observational difficulties from errors peculiar to social inquiry. Respondents may conceal information to evade taxation, resist official interference, or mislead competitors. Mechanical processing cannot eliminate these defects in the original evidence. Nor does aggregation automatically cure them: opposing errors might accidentally yield a correct total without making the observation reliable.

Yet any statement that errors cancel one another has to be proved.

This shifts the burden of proof onto users who assume that large aggregates or repeated observations neutralize uncertainty. Recent statistics are not necessarily better either: technological change, new products, discriminatory exchange rates, and political manipulation can make an economy harder to describe. Morgenstern further distinguishes excessive numerical precision from “functional speciousness”: an official exchange rate can be recorded perfectly and still misrepresent the rates at which transactions actually occur.

The discussion of unemployment shows why statistical uncertainty is also conceptual. Frictional unemployment, withdrawal from the labor force, movement back to farms, and occupational displacement complicate the category being counted. Political incentives compound those difficulties. Discrepancies between American statistical series make confident interpretations of small monthly changes especially questionable. National income then brings the conceptual and observational problems together:

The two areas are interdependent, since nothing can be measured for which there exist no good concepts, and concepts, no matter how precise, are of little practical value if the corresponding measurements cannot be performed.

An aggregate combines sectors with markedly different evidential foundations. Drawing on Simon Kuznets’s assessment of 520 component cells, Morgenstern contrasts relatively well-documented manufacturing and utilities with much less reliable construction, real estate, and services. He treats errors of 5 or 10 percent as economically enormous, while criticizing the official practice of leaving reliability judgments to individual users. Agencies possessing the resources to compile the figures cannot reasonably transfer responsibility for assessing their uncertainty to readers.

International comparisons intensify these problems because both concepts and evidence vary across economies. Subsistence production and uncertain population counts obstruct apparently straightforward per-capita comparisons. Soviet statistics introduce additional problems of falsification and multiple counting, illustrated through criticisms by the Soviet economist S. G. Strumilin. Morgenstern allows that broad differences between countries can be informative; he rejects the assumption that such differences are reliably expressible in several significant digits.

Growth rates receive the article’s most sustained scrutiny. They inherit uncertainty from national-product estimates and from the prices used to deflate them. Hypothetical calculations show that modest errors in two successive totals can radically change the magnitude, or even the sign, of the resulting rate. The tables illustrate sensitivity to possible errors rather than establish a measured probability distribution. Their force lies in exposing what confident interpretations silently assume:

And it cannot be assumed without further proof that the errors remain constant over time, that they change uniformly over time, and that the signs of the errors never reverse themselves.

Later revisions and discretionary base-year selection add further instability. A reported rate depends partly on when it is calculated and which starting point is chosen. Qualitative evidence and longer periods can support judgments about development, but comparisons between structurally different countries require particular caution. Moreover, identical aggregate growth can conceal contrasting productive changes: additional swimming pools and automobiles differ from investment in machine tools and power plants. Morgenstern proposes more purpose-specific “power indexes” to distinguish developments that a single total obscures.

The conclusion joins institutional reform to methodological revision. Substantially falsified data should be discarded; honestly collected but inherently uncertain data require changes in the theories applied to them. This does not mean postponing theory until exhaustive description becomes possible. Theory remains a necessary, revisable commitment. The practical priority is to end the presentation of economic statistics as faultless:

Perhaps the greatest step forward that can be taken, even at short notice, is to insist that economic statistics should not be released without an accompanying estimate of their error.

Even rough estimates would discipline both producers’ claims and users’ demands. The article’s enduring methodological point is that scientific economics requires explicit relations among concepts, measurement quality, and permissible inference—not merely more data or finer numerical detail.