Making Measuring Instruments

T h e A g e o f E c o n o m i c M e a s u r e m e n t A n n u a l S u p p l e m e n t to Volume 33 History of Political Economy Edited b y Judy L. and M a r y S. K l e i n M o r g a n D u k e University Press D u r h a m a n d L o n d o n 200,1 Perspective Making Measuring Instruments M a r y S. M o r g a n In the mid-nineteenth century, economists had many numbers but relatively few measurements; by the mid-twentieth century, they began to take for granted that they had measurements for most of the phenomena about which they theorized. This is the change that marks the impact of the age of measurement, and my interest in this prespective essay is to provide an understanding of what this change entailed. Out of many different economists' individual practical measuring projects around the late nineteenth and early twentieth centuries, there emerged a number of different kinds of measuring instruments, each carefully fashioned for economics. By looking carefully at the materials collected in this volume, we can discern that these instruments can be grouped according to the strategies or recipes that were followed in making such measuring instruments. As a historical development, first the instruments, then the strategies, served to connect the ambitions of economists seeking accurate scientific measures of their world with the practically necessary, sometimes mundane, but never trivial tasks of data collection. Correspondence may be addressed to Mary S. Morgan, Department of Economic History, London School of Economics, Houghton Street, London WC2 2AE, U.K.; e-mail: [email protected]. My thanks go to all the members of and contributors to the measurement workshops at the University of Amsterdam and the London School of Economics over the last few years and to the hard-working contributors to the HOPE workshop and this volume. I thank particularly Marcel Boumans, Harro Maas, and Hasok Chang for guiding my thinking about measurement issues. I acknowledge support from the British Academy for funding this work. 236 Mary S. Morgan Both measuring instruments and measurement strategies are a consequence of the age of measurement, not a prerequisite for it. Making measurements of economic entities requires instruments of a certain kind. The second broad historical thesis suggested by the essays in this volume is that the measuring instruments that economists built in this period were also analytical devices. These instruments embodied frameworks and techniques to turn observations into measurements and to organize the empirical data into particular types of arrangements so that economic phenomena might be arrayed before us. On one view, we can see this as a historical process of establishing facts about phenomena, paralleling the epistemological interests of Bogen and Woodward (1988).' On another view, this process can be seen, ultimately, as one of creating new categories of phenomena, paralleling the historical interests of Theodore Porter ([1994] 1996). I begin in the former tradition and will return to the latter interpretation. Measuring Instruments It is a conventional, and surely uncontested, claim that taking measurements requires instruments of measurement. From our childhood use of fingers and rulers to the technicalities of clocks and Geiger counters, we rely on tools to help us measure our world. Economics is no different, except perhaps in being relatively slow to develop such measuring instruments. The discipline did not develop such devices until the late nineteenth and early twentieth centuries, whereas there had been a dramatic increase in the number and range of measuring devices in the natural sciences in the seventeenth and eighteenth centuries (Heilbron 1990). Economists of the late nineteenth century wanted to measure all sorts of things, from the close-up single choice of an individual to the aggregate price level changes in the economy as a whole. But they had few instruments that would allow them to do these things. Whereas some might suppose that economic measurement is merely counting "what is there," a similarly naive view would have us think that x-ray machines merely look through our flesh to reveal our bones. We don't see a macroeconomy, nor a consumer price index, nor an individual choice decision, hence fashioning measuring instruments in economics has been, in part, 1. Within this framework, the problem of differentiating facts from artifacts, and the role of measuring instruments in this process, are particularlyfinelytreated in Franklin 1986. Making Measuring Instruments 237 a matter of developing ways of observing the economy. It is consistent with this to note, from the history of science, that the design and use of measuring instruments are often aligned to experimental investigations.2 Our analogy continues with the advanced medical scanners, instruments of investigation that are designed and programmed to enable us to "see" certain things inside us by picking them out in a particular way. The ways in which the economic body is investigated and data are collected, categorized, analyzed, reduced, and reassembled amount to a set of experimental interventions—not in the economic process itself, but rather in the information collected from that process. Economic observations must not only be registered but also converted into measurements, and converted in ways which serve particular theoretical or empirical or bureaucratic purposes. All this is to say that modern economics has rulers that are as complicated and as technical in design as those of modern medical science. Economic measurement requires counting and arithmetic on a grand scale according to certain procedures. Of course, those procedures involve bureaucratic regimes, clerical workers, automated information systems, and so forth. But they also require sets of instructions, provided, as Marcel Boumans (1999) has recently suggested, by certain mathematical models, or parts of models, that we can think of as constituting measuring instruments in economics. Using this important insight here, we can interpret the mathematical formulae, statistical formulae, and accounting rules, as well as the rules of data elicitation and manipulation, described in the essays in this volume in terms of measuring instruments. The appropriateness of such a labeling lies in the fitness and fruitfulness of the historical reinterpretations it prompts. The terminology and concept of measuring instrument fit happily onto the work of several late-nineteenth-century masters of measurement. Marcel Boumans (this volume) uses his earlier insight to understand Fisher's index number formulae; Harro Maas (this volume) shows how William Stanley Jevons can be understood as using a "virtual balance"; and Franck Jovanovic and Philippe Le Gall (this volume) recount how 2. See for example, Chang 2001. Heilbron (1990, 6) suggests that the early scientific instruments can be labeled either as "measurers" (for example, the barometer) or as "explorers" (such as the air pump) which "produced artificial phenomena for demonstration, investigation, and measurement." This distinction might be difficult to maintain for instruments made by economists in the age of economic measurement, where the relation between these aspects of measuring instruments is very close. 238 Mary S. Morgan instrument terminology was explicitly used by Lucien March to describe his "business barometers." When such economists wanted to solve a new measurement problem, they adapted, or fashioned anew, an instrument to make those measurements. These measuring instruments can now be grouped and typed according to certain general features. For example, this is a period when many different index number formulae are developed, each one offering a differentiated measuring instrument for a particular purpose. But we can also see that all index numbers follow a generic recipe or strategy of "weighted averages." For another example: we can understand how the umbrella strategy of "using causes to make measurements" can describe the various ways Jevons used assumptions about causes to make measurement possible, although the different ways might be thought of as different measuring instruments. Once such instruments are forged, they tend to be taken for granted, and once strategies are recognized, they provide recipes for new instruments. Each individual instrument has its own history, but the sum total of the age of measurement is the emergence of strategies for making such instruments. Strategies for Making Measurements Using Causes to Bootstrap Measurements One strategy of measurement is to pay attention to causes. Since the 1920s or 1930s, econometrics has focused its energy on building models to represent relations in the economy. It then uses these models to establish which are the relevant causal factors and to measure the relevant strength of those factors in explaining certain economic phenomena represented in the models. This approach relies both on techniques, developed only in the twentieth century, to measure causal relations in statistical data and on the already established measurements of the variables in the relationships (see Morgan 1997). Econometricians practice a kind of indirect measurement of the causal connections in relationships based on their theoretical preferences and what are perceived, at best, as direct measures (or at worst proxies for those) of the elements to be related. But we find the role of causes was rather different in the measuring instruments of earlier times. Whereas twentieth-century econometricians assume measurements to access causes, nineteenth-century economists interested in measuring assumed causes to make measurements. Making Measuring Instruments 239 The problem as understood from Mill's time had been how to deal with multiple causes, some permanent and some accidental, and to pick out the effects of single causes from the multitude. Though it appeared possible in thought experiments to consider man as driven by a single motive of wealth acquisition, attempts to isolate and make measurements of single, even constant, causes in such circumstances were fraught with difficulty. Simple averages, the main statistical technique of the day, were thought to work on the assumption of many small disturbing causes which cancel out in the average, as in measurement errors. Sometimes these averages worked well, as in Jevons's measurement of the wear on coins discussed by Sandra Peart (this volume). Other times, as in Jevons's investigation of poverty, they were inappropriate—in such situations, often found in economics, each item involved in the aggregate is subject to many causes, all of them valid, and none necessarily small or assumed to cancel. We can also see these difficulties of multiple causes in Adam Smith's thinking about the value of money (in Kevin Hoover and Michael Dowell's discussion in this volume) and in Jevons's work on the same topic (as discussed by Hoover and Dowell, Peart, and Maas all in this volume). The solution used by Jevons in his study of gold can be interpreted as assuming a cause (namely, variation in the supply of gold, which alters all prices equally) in order to bootstrap a measure of the change in the value of gold money, whereas later economists might start from measurements of money stock and prices and use these to measure the parameter relating the two as a test of the quantity theory. Jevons assumed that there was a reciprocal relationship between the value of gold money and the general level of prices, that the supply of gold would determine the general level of prices, and that other causes of individual price changes reflected in the data on individual prices were independent and would cancel out. He could then interpret the average change as attributable to variations in gold supply and thus provide a measure of the change in the value of gold: the assumptions about causes provided the basis for the measurement of the element of interest, namely, the change in the value of gold. The kind of circularity inherent here, that of inferring a quantity on the basis of assuming a lawlike relation, is found in many fields, and need not necessarily be dismissed on the basis of its circularity, for as Chang (1995) argues, such circularity can be fruitful in concept building and inferences may mesh with other measurements obtained elsewhere. 240 Mary S. Morgan The circularity would only become dangerous if Jevons's measure of the change in the value of gold was then followed by the kind of work undertaken in modern econometric practice in which such a measurement is used in a test of the quantity theory. It is to Fisher's credit that his work on this problem in the early twentieth century steered away from this measurement circularity in two connected steps: first, he represented the relation between money and prices in an equation of exchange which incorporated two other elements, velocity and trade; and second, he constructed independent measurements of all four elements in the equation. Nevertheless, using causal assumptions to measure money changes remained fraught during the period from Jevons through the twentieth century. Thomas Humphrey's analysis (this volume) of the 1920s debate between Federal Reserve Board economists, who believed in the real bills doctrine of money, and the quantity theorists shows how intimately causes, concept definitions, and measurement structures (including the rules of the institution) are all bound together. The inextricable mixture of definitional questions and the use of causal claims with measurements represented in this episode might be considered common for the period between Jevons's usage of causes in constructing measurements and the methods generally accepted after the 1940s for establishing causal relations (assuming existing measurements of variables) which came with econometrics. Adding with Weights to Make Index Numbers Index number formulae conceived as measuring instruments are based on the strategy of aggregating in a way that allows each individual element to be assigned its due weight in the whole. Such a "weighted average" strategy provides a solution to a general problem in economics, namely that many concepts refer to aggregates of things which may be considered homogeneous in the dimension of prices or money value, but are nonhomogeneous in another dimension, namely amounts consumed or produced. The solution is to use weights to overcome the problem of how to average in a manner that takes account of both amounts and values. The weighted averaging strategy of measurement that forms the basis of index numbers was developed in the late nineteenth century and became well accepted by the early twentieth century, although the problem Making Measuring Instruments 241 of designing good index numbers is far from trivial and vibrant arguments remain over which particular design should be used in which circumstance. For example, the arguments recounted by Spencer Banzhaf (this volume) are concerned with the problem of taking into account a third dimension, namely quality, alongside quantity in the price index. In his account we see arguments from microeconomic theorists that index numbers should measure constant utility equivalents and complaints from aggrieved consumers (during times of price rises) that official indices did not measure price changes accurately because of hidden quality changes. The hopes of both groups were defeated by the practical problems about how to weight items to take account of quality changes— despite the advice of academics on committees of inquiry and the willingness of bureaucrats, who collect the data and construct the measurements, to take these problems seriously. Measurement strategies, and the particular measuring instruments they spawn, often have an inherent design tension between satisfying consistency requirements and providing a measure of a concept that coheres with features of the empirical world. Instruments often embody both theoretical and empirical requirements in their design, creating two forms of internal consistency which ought to be met. Both these may place strain on the design and use of the instrument, a strain exacerbated by the desire for coherence. For example, it is well known that there is no perfect index number, just a very large set of possible ones. Some satisfy particular criteria from economic theory, others satisfy criteria which come from data rules. But the instrument design also needs to mediate between these two levels if it is to provide coherent measurements that link the empirical to the theoretical levels. Boumans's detailed examination of Fisher's work (this volume) shows how, during his work on index number design, Fisher moved from a design focus on theoretical consistency, regarded as mathematical features of the index formulae, to a realization that since index numbers could not meet all such criteria at once, his tests would be better interpreted as ones of coherence—how well a particular index number captured the relevant theoretical concept in a way that minimized distortion of the aspect of the empirical world being measured. As Roy Weintraub (this volume) reminds us, mathematics connects to the other sciences in different ways as its image changes. The mathematical virtues of truth and exactness might sometimes be connected with measurement virtues of accuracy and precision, while at other times they might cut loose from 242 Mary S. Morgan them. Boumans (this volume) uses the history of mathematics to analyze Fisher's changing understanding of measurement: theory dictates the nature of possible measurement assuming the theoretical picture of the world is correct, but that world itself may be differently arranged. A parallel example comes from the early work on the "identification problem" in econometrics. Econometricians of the 1920s and 1930s were adamant that in order to measure the parameters in relationships, not only did the mathematical structure representing those relations need to have certain features, but the statistical data needed to have variability over the same features (see Morgan 1990). The issue of coherence between these two sets of conditions lay at the heart of the 1930s "pitfalls" debate.3 The problem of coherence between theory and the world raises dangers for any measurement strategy. Using Properties of the Balance The strategy of using a balance for measurement purposes should come as no surprise: that is what a balance is for, and as Peart (this volume) shows, Jevons used a physical balance to measure the wear on coins. What is unusual in economics is that the balance is used not as a physical measuring instrument, but as a virtual instrument. We can see this in Maas's analysis (this volume) of Jevons's work both on the value of gold and on the measurement of individual feelings. In the former case, the balance analogy is used in a specific argument to structure the measurement problem and so to provide a solution to the problem of measuring the value of gold: here the measurement involves a weighing up procedure, and an averaging procedure, both of which, in their detail, depend on the properties of the balance. The measurement of feelings applies the idea of the balance to the individual and her perceptions of utility: here the mind is portrayed as the balance, the mind registers both moments of equality and the points at which the balance between two things tips in one direction. Jevons describes a procedure of measuring sufficient for the individual to decide action; there is a judgment of equality or inequality and no numbers are needed. Just as an hourglass can measure the passage of an hour without telling the time of day, the balance 3. See Hendry and Morgan 1995 for extracts from the debate and earlier papers on identification along with a commentary on these matters. The insights of those years were temporarily lost as "identification" came to be understood as a characteristic of the mathematical structure alone. Making Measuring Instruments 243 can judge the comparative heaviness of two things without telling their weights. Accounting, Wholes and Parts Accounting principles provide another generic strategy for economic measurement. Identities, accounting coventions and arithmetic identities defined by theory, hold things together and provide an important rule constraining the measured parts to fit together in a particular way: if two sides of the identity don't equate, something is missing or overcounted. Balancing requirements within accounting, either by the imposition of an equation or by using double-entry systems, provide another constraint available where theoretical structures stress equivalences of inputs and outputs as in the Leontief system (discussed by Martin Kohli in this volume), or the overall total equivalents of expenditure, output and income as in Richard Stone's national income accounts (discussed by Flavio Comim in this volume). These accounting elements provide both a strength and a weakness. Their summing and balancing requirements seem to ensure accuracy, whereas in reality they ensure only a type of precision. Everything that has to balance or add up must do so: precision, but this might be achieved at the cost of only partial coverage of the aggregate: inaccuracy. The strategy rewards consistency at the empirical and theoretical levels but shortcomings of coherence between the two may be neglected. Another well-known advantage of the accounting strategy is that it is flexible to many different theoretical accounts of the economy, to different economic ideologies and to many different levels of application (see Morgan 2002). National income counting predated Keynesian economics and although it later became strongly associated with Keynesian theory, it has not, historically, been tied irredeemably to Keynesian concepts. Input-output accounts can be used within a Marxian economics framework and a planned economy accounting system as well as with a regulated or managed market economy. They can be constructed for fineor coarse-grained measurement and analysis, for use with national income accounts or as part of a national budgeting system. This flexibility points to the fact that whatever accounting scheme is adopted is largely a matter of choice and is historically contingent, determined by the requirements of governments in war, the local economists hired to make 244 Mary S. Morgan the accounts, and so forth. Thus, in the same period, we see the MeadeStone accounts developed to work out wartime expenditures in Britain; Leontiefs input-output accounts used to predict future manpower levels and as a check on the national accounts developed by Simon Kuznets in the United States (see Comim's and Kohli's articles in this volume); and a particular form of national budgeting developed in Norway (see Bjerkholt [1998] 2000). The choice becomes a conventional one once national income accounts have been standardized under the auspices of international economic agencies in the later twentieth century and those agencies used their power to insist on their own measuring instruments. Related to these accounting measures are strategies of measuring wholes and parts. One of the oldest and well-measured wholes is the population, with censuses becoming widely established just after 1800. The population might be considered something like a natural whole: difficult though it might be to count every person, in principle it is relatively easy to define the whole. Sybilla Nikolow's essay (this volume) shows how far early measurers in the German cameralist tradition could emphasize the importance of such population and geographical measures of the whole by using them to work out ratios of economic power and comparing different national results. With the twentieth-century development of macroeconomics, attempts to measure "the economy" taken as a unit whole (rather than an aggregate) proved much more difficult. Conventional definitions tend to cover only things which pass through the market place (including the government sector) and are thus accounted for in monetary terms. But this leaves out all sorts of economic activity ranging from neighborly exchanges to unpaid housework. Immediately, then, the strategy of what counts as part of the whole excludes some things and includes others in the count. The strategy of extending the accounts to wider wholes and breaking them down into narrower categories, or into different sets of parts, is associated with the accounting ideal along with a desire for classification and naming. But each such redefinition of wholes and reclassification of parts results in a complete revision of the measurement task. We can see this equally well in Comim's discussion (this volume) of Stone's desire to extend the accounts while standardizing them and in Kohli's discussion (this volume) of Wassily Leontief's difficulties in stretching the input-output accounts to focus beyond production. Making Measuring Instruments 245 Constructing Indicators The measurement strategy which makes indicators to inscribe the path of the economy over time has often been derided as hopelessly empiricist, subject to "measurement without theory" jibes. Yet the creator of the French economic barometer, Lucien March, considered his task to be the extraction of regularities out of a complicated pattern of information, in much the same way as others had tried to extract measurements on one common cause out of the whole set of interlocking multiple causes. March referred to his indicators as "instruments of observation and analysis" (Jovanovic and Le Gall this volume). He characterized the numbers the barometer inscribed as "conventions," in the same way, we might add, as numbers registered by thermometers are conventions (either degrees Fahrenheit or Celsius), but nevertheless valuable for comparisons, and once the convention is known, for indicating absolutes. Although called "index numbers" by March, these business barometer indicator measures had little in the way of principles dictating their construction compared to the weighted averages measures referred to above. Nor were they like the indicator diagrams produced in engineering and in psychology in which the indicators are a directly inscribed outcome of some physical process or some physiological reaction (see Brain and Wise 1999). The indicator diagrams of economics, seen in the Manchester Statistical Society's pursuit of a "weather glass" to indicate the "dark days" of the trade cycle (see Judy Klein's essay in this volume) and in March's business barometer of the early twentieth century, are both forerunners of the business cycle measurements of the 1930s. These measurements are the results of time-series data, carefully filtered and amalgamated according to different rules. Some rules are conventional (for example, detrending), some rules are driven by theories (for example, about which data series constitute the phenomena), and some are more ad hoc (for example, based on correlation measures). The indicators are then charted so as to trace out the movements of economic activity over time. This is a measurement strategy which depends heavily on quantitative techniques yet is barely bound by rules from accounting, balancing, or weighting conventions and makes little call on principles deriving from theory or knowledge of causes. More than other strategies, this strategy makes measurements without the benefit of any blueprint other than the techniques supplied by earlier makers of business barometers. This is ironic, for this measuring instrument is one of the ' 246 Mary S. Morgan few in economics named after an existing scientific measuring instrument which is theory bound. Though not shaped by strong principles, such time-series indicators have become one of the more well-used measuring instruments in twentieth-century economic life. Just as those who own a barometer read it on their way out the door, business people, investors, government ministers, and their economic advisors check the leading and lagging indicators of the economy to see where it is heading. Social Surveys Social surveys are perhaps the most easily conceived of as a strategy for measurement, for they embed the sampling principles and techniques from statistics alongside an open-ended, flexible observation and counting method which can be applied across a range of topics, depending on the historical context in which this demand for measurement is made and met. As Bradley Bateman (this volume) suggests, there were many examples of survey data being collected in economics in the late nineteenth century—on budgets, on insurance systems, on poverty and wages, and so forth. Though the survey went out of fashion in economics during the mid-twentieth century, replaced by the strategy of causal measurement in econometrics, it has since returned as an essential element of econometrics itself. Much of modern microeconometrics and microsimulation (which perhaps ought to be treated as a measurement strategy on its own) relies on data gained through surveys, and econometric techniques have had to be adapted to the measurements generated by such instruments. The Components of Strategies The development and adoption of strategies of measurement that I describe here may have been overlooked, but they have been no less effective for all that. Historians of economics have focused on individual cases of measurement history, and, taken individually, what we see is the development of specific measuring instruments, each of which might be described as a mixture of the conventional and the ad hoc. It is only when these developments are viewed together that we can discern and describe a history in terms of a few generic types of measuring instruments. The historical change that marks the age of measurement is not therefore to be understood just as a massive increase in the collection Making Measuring Instruments 247 of numbers but as a set of processes that only begin with counting, and then go on to analyze, reduce, aggregate, and reform our measures of the economy in ways unthinkable in the early nineteenth century. It is these processes, taken together, and with hindsight, that reveal themselves as a set of measurement strategies. What do I mean by the term strategy here? It may be defined as a method for making, doing, or accomplishing something. Its synonyms suggest an idea, plan, or design: a pleasing combination of practical and ideal, yet the terms are still too vague to describe what was happening in economic measurement during the period from the late nineteenth to the mid-twentieth century. Each economic measuring instrument can be understood as involving three elements, namely, of principle, of technique, and of judgment; but there is also an overall element of design which I call a strategy. A particular strategy constrains the choices and the combinations of the three elements, and these elements in turn shape the way individual measuring instruments are constructed and so the measurements that are made in economics. For example, we can characterize both accounting principles and the equations we get from economic theory as the kinds of principles or rules which are embodied in measuring instruments. Techniques also come in various forms: techniques for counting and calculating, for cleaning data of outliers and "noise," for choosing samples, for meshing observations, and so forth. Judgment plays a role in many places, but particularly where design decisions have to be made about classification, exclusion, and inclusion. These judgments or decisions are often determined by the purpose for which the measurement is made. But judgment, techniques, and principles are not freely chosen: in the case of each measuring instrument, they hang together within an overall strategy for making instruments of that particular kind. Economists with as diverse aims and epistemological beliefs as Irving Fisher and Lucien March could both, at the beginning of the twentieth century, write seriously about the design of measuring instruments, their construction, and the criteria they should meet. Both indeed constructed "index number" measuring devices, but with very different principles and techniques. Because they were following different strategies, they constructed different kinds of instruments, and made different measurements. In contrast, the individual designs and constructions developed by Stone and Leontief can be recognized in terms of the same generic strategy—although, since Stone and Leontief judged different categories ' : I , , • 248 Mary S. Morgan to be important, their measuring instruments were constructed to measure different things. Fisher, March, Stone, Leontief, and many others shared a strong commitment to the measurement ideal, and their measuring work forms the basis for now well-accepted measuring instruments.4 The Life of Measurements s I Si I have painted an account of measurement strategies as epistemological instruments—to the neglect of the social, economic, and bureaucratic realms in which their history lives (see the essays by Klein and Porter in this volume). Yet the change wrought by the age of measurement is at least as evident in the way that we now habitually accept that the measurements provided by these instruments give us reliable representations of what is happening in the economy, and provide an impetus on which we as individuals, and on which our governments, act. The importance of numbers to social knowledge and action is made most evident in Bradley Bateman's essay. But Porter (this volume) emphasizes another aspect of the importance of numbers and reminds us of one of his cherished themes when he suggests that only in the period after 1900 did economists "systematically construct and analyse new entities through measurement." The measuring instruments discussed here are critical to this, for as Porter ([1994] 1996, 36) argues, "standardized quantitative rules have been almost as fertile as standard experiments and massproduced instruments in the making of new things." Measurement, maybe even more than theory, contributes to the making of new ontic furniture for the economic world. Porter reminds us, in his 1994 essay, that the history of science is riven with examples in which theoretical disagreements over the nature of things or lack of knowledge of causes failed to impede measurement; that quantities often displace concepts; and that quantitative "laws" have often been taken as evidence for the reality of elements in the laws. Though the wealth of the nation or the economic power of the national economy was an early category for discussion, it was only after 1950 that the Gross National Product (GNP) 4. One indication of the importance the profession attaches to these measurement strategies is to note that nine master craftsmen of measurement have been awarded Nobel prizes in economics: for the causal account (econometrics) strategy, by Jan Tinbergen, Ragnar Frisch, Trygve Haavelmo and Lawrence Klein; for the accounting strategy by Simon Kuznets, Richard Stone, and Wassily Leontief; and most recently to James Heckman and Daniel McFadden, who fashioned the econometric techniques which enabled economists to use survey data effectively. Making Measuring Instruments 249 moved from being a category of economic analysis to being a number of everyday concern to government. Its measurement, tied at some point to a particular theory of macroeconomics, broke away from its theoretical roots to become a theory-neutral (though not value-neutral) measure of the economy's progress and health. The monthly change in the consumer price index (CPI), like the daily temperature, is a regularly reported number in our newspapers and influences all sorts of decisions at all sorts of levels. Again, almost regardless of changing knowledge and theories about what causes the CPI to change, it is treated as a real quantity by the participants in an economy. Other categories of measurement are more limited in their circulation, being "real" for certain economists, officials, and businesses, but otherwise remaining technical terms that hardly touch the general consciousness. As we have seen, these "real" quantities are constructed with measuring instruments that have become conventional quantitative tools of analysis yet whose starting point and reasoned development has often since been forgotten. Like thermometers, we need only to know how to use such instruments, not necessarily how and why they work. These measuring instruments report "facts" about phenomena for us, but the phenomena they describe have been constructed on instruments fashioned long ago from the experiments and choices made by scientists in the past (see Chang 2001). There is no "right" set of rules for GNP construction, one that fits naturally to our economic world. Like a new glove, awkward at the start, a new measurement gradually becomes comfortable, though it may never fit tightly. We choose our accounting conventions, and only after the passage of time does a measured concept like the GNP conform to our other knowledge and come to seem "real." If we do our own taxes, we make the same transition: tax authorities' accounting conventions rarely fit each human experience—our experiences have to be made to fit their conventions. This same process happens with economic measurement at other scales and with instruments other than accounting. Economic measuring instruments are not devices given by God for us to reckon his own preordained economy; they are the inventions of economists and we fit ourselves to their measures. As Werner, one of Goethe's fictional characters, remarked of double entry bookkeeping, "It is among the finest inventions of the human mind; every prudent master of a house should introduce it into his economy." The character of Wilhelm replied, "You begin with the form, as if it were 250 Mary S. Morgan matter: you businessmen commonly forget, in your additions and balancings, what the proper total of life is."5 References Bjerkholt, O. [1998] 2000. Interaction between Model Builders and Policy Makers in the Norwegian Tradition. In Empirical Models and Policy-Making: Interaction and Institutions, edited by F. A. G. den Butter and M. S. Morgan. London: Routledge. Bogen James, and James Woodward. 1988. Saving the Phenomena. Philosophical Review 97.3:303-52. Boumans, M. 1999. Representation and Stability in Testing and Measuring Rational Expectations. Journal of Economic Methodology 6:381^-01. Brain, Robert M., and M. Norton Wise. 1999. Muscles and Engines: Indicator Diagrams in Helmholtz's Physiology. In The Science Studies Reader, edited by M. Bagioli. New York: Routledge. Chang, H. 1995. Circularity and Reliabilty in Measurement. Perspectives on Science 3.2:153-72. . 2001. Spirit, Air, and Quicksilver: The Search for the "Real" Scale of Temperature. Historical Studies in the Physical and Biological Sciences (forthcoming). Franklin, A. 1986. The Neglect of Experiment. Cambridge: Cambridge University Press. Heilbron, J. L. 1990. Introductory Essay. In The Quantifying Spirit in the 18th Century, edited by T. Frangsmyr, J. L. Heilbron, and R. E. Rider, 1-24. Berkeley: University of California Press. Hendry, D. F, and M. S. Morgan. 1995. The Foundations of Econometric Analysis. Cambridge: Cambridge University Press. Jackson, M. W. [1994] 1996. Natural and Artificial Budgets: Accounting for Goethe's Economy Of Nature." In Accounting and Science: Natural Inquiry and Commercial Reason, edited by M. Power, 57-81. Cambridge: Cambridge University Press. Morgan, M. S. 1990. The History of Econometric Ideas. Cambridge: Cambridge University Press. . 1997. Searching for Causal Relations in Economic Statistics: Reflections from History. In Causality in Crisis: The New Debate about Causal Structures, edited by V. McKim and S. Turner. South Bend, Ind.: University of Notre Dame Press. 5.1 owe this wonderfully apt interchange to Jackson ([1994] 1996). The source is Goethe's Wilhelm Meisters Lehrjahre of 1795. Making Measuring Instruments 251 . 2002. Economics. In Modern Social Sciences, edited by T. H. Porter and D. Ross. Vol. 7 of The Cambridge History of Science. Cambridge: Cambridge University Press. Porter, T. H. [1994] 1996. Making Things Qualitative. In Accounting and Science: Natural Inquiry and Commercial Reason, edited by M. Power, 36-56. New York: Cambridge University Press.