An essay on industrial classifications

(Article by Bernard Guibert, Jean Laganier and Michel Volle in Économie et statistique No 20, February 1971)

French version: Essai sur les nomenclatures industrielles.

There can be no economic analysis without a classification. Only a classification can give precise enough meaning to the terms that crop up so often in economic reports - "textile industry", "furniture", "steel industry" and the rest. Classifications play an absolutely crucial role, but they tend to be dismissed as tedious. They consist of tiresome lists with only the occasional intriguing oddity to break the boredom. A classification specialist is seen as a real technology geek, and has to be exactly that to answer the seemingly hair-splitting questions (s)he is faced with every day: should the manufacture of plastic footwear come under footwear manufacture or under manufacture of plastic products? What is the distinction between shipbuilding and the building of pleasure boats? Should joinery be classed as manufacture of wood and wood products or as building construction?

The authors of this essay (1), who have used industrial nomenclatures on a daily basis either at INSEE or at the French Ministry of Industrial and Scientific Development, attempt to explain and then to overcome this prejudice. Their presentation, which aims to identify the principles on which present and past classifications were built, knowingly or otherwise, would probably not be accepted by other specialists without adjustment. It does succeed, however, in turning a supposedly disagreeable topic into an accessible, even fascinating, one).

*     *

According to Heidegger, "for someone who wears glasses - a device which, depending on the distance, may be so close to him that it 'sits on his nose' - the device is, in terms of his environment, more remote than the picture on the opposite wall. It is so close that it usually passes unnoticed." Commenting on this statement, Bourdieu says: It is the same ethnocentrism which tends to regard as realistic a depiction of reality that owes its "objectivity" not to its concordance with the actual reality of things (since that reality is only ever delivered by way of socially-conditioned forms of perception), but to the fact that the rules defining its syntax in its social usage agree with a social definition of the objective view of the world. By qualifying as realistic certain depictions of reality (photography, for example), society is merely confirming its own tautological certainty that an image of reality which agrees with its depiction of objectivity is truly objective [1].

This explains why the problems inherent in classifications appear superfluous, boring and pointless. It is not because they are unimportant; on the contrary, it is precisely because they are fundamental. An economist - if we may use this image - is not interested in the glasses through which (s)he sees the economy; (s)he is, however, greatly interested in what (s)he sees. If we want to see the glasses we are wearing, we first have to take them off, and that blurs our vision. Similarly, discussions of classifications lead us to view as fragile, malleable and ultimately questionable those aggregates whose robustness we used to take for granted. Dividing-lines which used to be sharp become uncomfortably fuzzy.

Well, for once we are asking the reader to examine those glasses carefully. What do the classifications look like? How are they constructed? Which ones were used in the past, and what is happening today?

1 - What is a classification?

It is not a question of linking consequences, but of comparing and isolating, of analysing, of matching and pigeon-holing concrete contents; there is nothing more tentative, nothing more empirical (superficially, at least) than the process of establishing an order of things; nothing that demands a sharper eye or a surer, better-articulated language; nothing that more insistently requires that one allow oneself to be carried along by the proliferation of qualities and forms. And yet an eye not consciously prepared may well group together certain similar figures and distinguish between others on the basis of such and such a difference: in fact, there is no similitude and no distinction, even for the wholly untrained perception, that is not the result of a precise operation and of the application of a prior criterion.
Michel Foucault, Words and Things

I reflected that Argos and I belonged to separate worlds; I reflected that our perceptions were identical, but that Argos combined them differently and made other objects out of them.
Jorge-Luis Borges, The Immortal

Although we have to start with a formal definition, even to distinguish a classification from a non-classification, the definition will not make sense until we have examined the process of construction and the habitual use of classifications.

A set of nested divisions...

Let us first look at a product classification, the customs nomenclature, taking as our guide not the method adopted by its authors but the appearance of an ordinary page of this document.

It contains a list of basic product classes: "cotton coats for women and girls", "machine screws for wood, threaded, in bulk, shank diameter not more than 6 mm", "B2 vitamins", etc.

There are also groupings described in more general terms: "Women's, girls' and young children's outerwear", "Bolts and screws etc. and similar articles in cast iron, iron or steel, etc.", "Natural or synthetic provitamins and vitamins, etc.". These first-level groupings are themselves placed in second-level groupings: "Clothing and clothing accessories in fabric", "Cast iron, iron, steel", "Organic chemicals".

This products nomenclature thus looks in practice like a set of nested divisions (2). Everything is grouped from the bottom up to a top, unspecified level which we might call "products".

The nomenclature thus contains both highly aggregated levels and levels of detailed breakdown.

A numerical code is attached to each basic class and to each node of the nomenclature. The numbers of all headings at a particular level have the same number of digits (they are known as the "two-digit level", "four-digit level" etc.) and the number of a "node" identifies both the node itself and higher-level groupings which contain it. Code numbers are thus allocated "top down" from the top of the tree (3).

To summarise, a classification looks formally like a series of divisions nested within a set of elementary headings. The list of these elementary headings must itself be defined in such a way that one single item in the list corresponds to each physical object in the field of study: in other words, a last division, that of the set of units (4) in the field of study, must correspond to the denominations represented by the list of elementary headings.

From the general to the particular, or vice versa?

We now have an exclusively formal criterion for recognising easily whether or not a text is a classification. Yet this criterion tells us nothing about the significance of classifications: the reason for their existence and how they are constructed. The formal response answers only part of the question we are asking: what is a classification?

We can answer part of the question by looking at how the classifications are put together. We might think of constructing a classification "bottom up", i.e. starting with a list of "basic" denominations, building up a series of intermediate thought-objects and finally arriving at the aggregate defining the field of study as a whole; we could also take that aggregate directly and divide it up successively so as to arrive at a list of elementary headings so detailed that we need go no further.

These two approaches, though formally inverse, are used simultaneously in practice. This is worth comparing with the Aristotelian notion of categories defined either in extenso (i.e. by exhaustive enumeration of the units designated by the category) or in intenso (i.e. by specifying the properties common to the units). If we eliminate any trivia, we can see that the categories are never compiled either in extenso or in intenso, but according to a process which in effect incorporates both of these contradictory approaches.

The list of elementary headings is the tricky part of the nomenclature; it is here that the relationship with the concrete is closest.

This list is not a set of objects (i.e. industrial, animal or human-activity products etc., according to the domain in question), but a set of denominations: two objects ranked under the same denomination are regarded as equivalent from this point of view.

To be usable as a basis for a classification, these denominations must meet certain conditions:

• as we have seen, they must be a subdivision of the set of singular units;
• they must reflect the purpose of the study.

We know that no unit can be assimilated into another, since a sufficiently detailed examination would always lead us to perceive a difference. A denomination is constructed from properties common to the units it designates, regardless of the differences they may otherwise reveal, and we obtain a list of denominations which is longer or shorter depending on the degree of detail sought. The threshold must be set by usefulness criteria, i.e. by letting the purpose of the study determine the structure of the list of elementary headings.

Only a usefulness criterion enables us to say, for example, whether or not we should break down cars in a product list by color. It all depends on the purpose: the distinction may be of no interest to an economist but crucial to a sales specialist.

Moreover, the thought-objects covering the denominations are not reconstructed each time we embark on "naming" the units in a field of study. Most of such thought-objects are presented to us ready-made by the historical context in which we find ourselves, and above all the most important of them: the field of study itself (5). The compilation of a classification is thus inseparable from a point in time which supplies the definition of the object, the linguistic material, the purpose of the study - in fact more or less everything except the classification itself, which we have to compile on the basis of the given parameters.

Forms and the significance of forms

The unit, the physical object situated in the field of study, is formless when it comes to us. The intellectual comprehension of this object gives it a form enabling it to be adapted to the requirements of intelligent action.

Thus an object which is ostensibly simple, e.g. a piece of merchandise such as a car, may be seen from two different angles. It can be said to exist as a conglomerate of atoms, as a piece of matter, but the same can be said of any stone or mineral sample. Our society's observance of this object lends it an entirely different significance. The laws of fashion, aesthetics, comfort, the technical constraints it obeys according to the means available at the time, even the set of affective mechanisms it invokes - all of these turn the object into a sign, a phrase of a social language which is not expressed in words but is no less meaningful for that. So what can we say? Is a car a lump of atoms (itself a social construct - there is no getting away from this) or is it a sign? It is both, and the one no less than the other.

Similarly, an economic activity can be seen from two different angles. At the most "material" level, each economic activity can be represented by an input which is turned by some process into an output. Boards and nails, once transformed by skilled hammer blows, become a packing case. Similarly, a cheque validated by a signature becomes a means of payment.

We could, of course, try to reconstruct in our minds the way the economy works by staying at the level of a sprinkling of objects. It is clear, however, that there are better things to do than to record a multitude of particulate acts; it is even more clear that the economic act of signing a cheque has its weight in its social significance and not in its anodyne material appearance: who carries out this operation, for whom, and why? It is worth noting too that debt discharge is gradually losing its material back-up with the rise of credit cards. Thus a gap is emerging between social significance and subjective perception.

The same applies to the more modest concept of the manufacture of packing cases. Cases are made, of course, but they have been ordered, or at least the producer is counting on a demand; but the technique used, however primitive, has a history in our society. Thus the act of production, when seen in a social context, itself becomes a sign. The same would be true of a more technically complex act of production such as car manufacture, mentioned above.

The whole of the economy, in its apparent chaos of tangled operations, speaks to us in its own language. All these acts, an impressive mass of signifiers, contain something signified. But the signifier delivers what is signified only if it is perceived and received in the appropriate forms.

Compiling classifications is one of the methods by which our society has attempted to draw out these forms. Classifications are an attempt to trace, amongst the chaos of signifiers, those parts whose juxtaposition we hope will reveal what they signify.

Let us now apply these ideas to our actual experience.

Universality and univocity

We have seen that each singular unit must have a corresponding denomination in the elementary list: the classification must be universal in the field of study; only one denomination may correspond to this unit: the classification must be univocal.

These two requirements directly reflect the fact that each level of the classification must correspond to a division of the universe of singular units.

This requirement could be seen as trivial, but it is nevertheless interesting to give it some thought, and we will quote some examples to illustrate its significance.

An expression such as "Construction of motor vehicles and cycles" evokes in everyone a fairly vague image of a complex of establishments, machinery and workers whose activity, indistinctly grasped as being fairly complex, has its end purpose in the marketing of motor vehicles and cycles. If we sought to delimit this complex with any precision, however, we would be risking substantial variations in the boundaries of the domain according to the individual point of view. For an investor, the denomination would cover manufacturing; for a user it would also comprise maintenance and repairs. The use of a common nomenclature enables everyone - provided that it contains the denomination in question at some level - to apply a concept with an unambiguous definition. Thus, in the Economic Activities Nomenclature, the content of heading 26 (motor vehicles, cycles) is broken down over two pages. The headings of the three-digit nomenclature are [21]:

261. Construction of motor vehicles with thermal engines
262. Construction of bodies (coachwork), trailers and semi-trailers
263. Manufacture of parts, accessories and spares for motor vehicles
264. Repair of motor vehicles
265. Manufacture of spares and accessories for cycles and motorcycles
266. Manufacture of motorcycles and cycles
267. Repair of cycles and motorcycles
268. Demolition of motor vehicles

The content of the aggregate is very precise. Problems arise when we consider activities which appear to belong both here and elsewhere: should garages carrying out motor vehicle repairs come under heading 264 or under 743 (garages, service stations etc.)? If we want to arrive at a strict definition of the content of a heading which intuitively seemed clear, we are obliged to solve a number of boundary problems, to amass specifications and warnings. The classification becomes more complicated and more unwieldy, and experience shows that some questions always remain in suspense.

The choices made in delimiting aggregates have important consequences; we can try a small exercise by looking at the page in Short-term economic trends [15] with the textiles index. It is clear that if, when defining the "textile industry" heading, we had excluded knitted and crocheted goods and man-made textiles (e.g. placing one under "Manufacture of wearing apparel" and the other under "Manufacture of chemicals and chemical products" (6)), the trend in the textiles index as a whole would have been quite different.

The conveniently concise terms we use to designate branches or sectors of the economy ("furniture", "textiles" etc.) therefore make no sense until we know the classification under which they are defined. If we want to draw comparisons, particularly international ones, the use of different classifications leads us to commit errors of interpretation. Readjustment calculations are possible only if we have been able to define a common level and if the data collection was enumerated in such a way that we can describe the same domain in the classifications used.

These requirements for precise contours to be given to each aggregate, to ensure that bridges are put in place each time we are compelled to "look at" the same domain "via" two or more different classifications, are of obvious importance. All this is about quality, of course - the care given to the editing of the classification; it says nothing about what we are actually expecting of the aggregates we are intending to manufacture. It remains for us to determine the criteria which will guide us in building these aggregates.

What criteria?

When faced with a classification which is perfectly clear, perfectly structured as a classification, users may well reject it. Let us suppose that we have a classification in which the manufacture of furniture and the manufacture of motor vehicles come under the same heading. This would obviously be rejected. "This aggregate makes no sense", people would say. But why?

Another example: in the customs nomenclature of products [19], perambulators and push-chairs (87-13) are placed very close to assault tanks (87-08) because both are "vehicles". Ordinary pumps, petrol pumps and fuel-injection pumps for diesel engines, which have only their name in common, are lumped together under heading 84-10. These are classifications which seem "natural" to the authors of the customs nomenclature but are strongly rejected by the authors of the activities nomenclature [16].

As always, the authors of this paper are using the word "natural" to denote a set of operations or attitudes which seem to them to fit together automatically; yet the different points of view prove that nothing fits together automatically. The choices made in compiling the customs nomenclature stem from its administrative function: it was created so as to enable differential charges to be applied, and must therefore be drawn up in such a way that the risk of classification error is as small as possible. It thus makes sense to use a classification which is partially mnemotechnic ("all pumps taken together").

A tool for economic analysis

This example shows how the criteria used for compiling a classification are determined by its intended use. In the case of a classification whose main function is to be a tool for economic analysis, we can say that the more an aggregate covers a set of items which is coherent in its depiction of the economy, the more likely it is to be accepted by an economist. The more coherent the aggregate, the more easily the economist will be able to see it as a single object of analysis; the more (s)he will be able to see the object as a subject with its own autonomy, i.e. with a "behaviour" and with "reactions". If the economist is totally taken in by its representative quality - i.e. if (s)he regards it as the very image of reality - (s)he may go so far as to attribute to this construct the role of a "natural" subject which appears "spontaneously", "freely" in the field of perception of the "objective" observer. Thus we speak of the investment behaviour of "the steel industry", or the stockpiling behaviour of "the distributive trades".

A tool for action

Let us now consider a different point of view. The classification is not just a tool for economic analysis: it is also a tool for economic action. Heads of large companies, professional organisations, government demand a breakdown likely to produce the information they need in order to take action; since their action uses existing institutions as a tool, channel or object, they often want an "institutional" breakdown.

This "technocratic" approach, which is found as often in private organisations as in the public sector, is clearly sometimes at odds with the requirements of economic analysis per se.

The historical analysis we are about to undertake shows the extent to which the various portrayals have, in order to construct their objects and economic facts, perfected criteria and classifications which serve them well. We shall also see how, since it has become possible for the State and leading players to act on economic structures, classifications are partly influenced by the structure of institutions.

2 - The history of industrial classifications

The history of industrial classifications is inseparable from that of industrial statistics.

Industrial statistics in France - and probably in the rest of the world - date from 1669. Colbert then advised that the situation of the kingdom's factories be recorded in numerical terms [9].

This operation was successful only in the textile domain, which was by far the most important. Studying the information supplied posed no classification problems.

The physiocratic classification of 1788: raw materials

In 1788, after a century of industrial development unaccompanied by new statistical investigations, Mr de Tolosan, superintendent for commerce, exercised his office in implementing the project designed by Colbert. Expanding his own knowledge with the help of the archives of the various ministerial departments, he drew up a table of France's leading industries and accompanied this with an assessment of the products manufactured by each of them [9].

This first French industrial nomenclature is given to us in this document by the order and title of the headings. It provided the framework for industrial statistics until 1847 and therefore deserves to be looked at in some detail. Industry is placed under three main headings relating to the origins of the raw materials used: "Mineral products", "Vegetable products", "Animal products". I - Mineral products: Rock salt and sea salt. Earthenware, porcelain. Glass and glazing. Pig iron. Lead. Copper. Ironmongery, haberdashery. Silverware, jewellery.

II - Vegetable products: Stationery. Starch. Soap. Refined sugar. Tobacco. Hemp, linen, cotton, canvas and other fabrics. Linen, knitted and crocheted goods. Cotton, knitted and crocheted goods. Linen, lace. Hemp, linen, ropes, net, thread ribbons.

III - Animal products: Silk garments. Tapestry, furniture. Furskins, cured goods. Woollen fabrics. Serge. Camlet. Coarse cloth. Broadcloth. Woollen knitted and crocheted goods. Hats. Silks. Silk knitted and crocheted goods. Ribbons, blonde lace, gauze, trimmings.

This classification of industrial activities takes a "naturalistic" view of the economy in which the influence of physiocrats is tangible; human activity causes nature to "bear fruit", but the emphasis is more on nature and its fruitfulness than on the act which induces the fruit to come forth. The portrayal lies within a "natural resources" ideology.

The criterion is applied without mercy: what we today call the textile industry is cut in half: hemp, cotton and linen are on the vegetable side, while wool and silk are found amongst the animal products.

Note that the "furniture" heading did not mean the manufacture of furniture as it does today, but essentially the production of carpets, which were made of wool. It was therefore natural to place it amongst the "animal products".

This nomenclature might be thought to have reflected only Tolosan's personal viewpoint. Not so; it was taken over as it stood by Chaptal, who published an estimate of French industry based on imperial statistics (7) around 1812.

1833: monographs on establishments

The Statistique de la France was abolished in 1814; in 1833 the government reinstated the Statistique générale du royaume (General Statistics of the Kingdom), with the approval of the chambers and to the satisfaction of all enlightened souls ([9], p. XVIII). From 1839, efforts turned to extending the investigations to industry. The instructions given to the prefects are worth quoting:

The general table of licensed business operators of each department is to be closely scrutinised, and from this a list is to be extracted of producers, entrepreneurs and manufacturers whose establishments extend beyond the arts and crafts class and belong to the manufacturing industry by virtue of their nature, their scope or the value of their products…

Conduct a detailed survey in each district, following the ideas supplied by these documents, with the purpose of establishing, by numbers, the industrial output produced each year by each factory or mill.

Include, however, only establishments which occupy with their work at least a dozen workers, excluding those who employ a lesser number, which shall in general be included in the arts and crafts class whose exploration will not take place until a later occasion.

Gather the statistical data relating to industrial establishments either by asking the owners or directors or, if information obtained from them is lacking, by carrying out assessments on the basis of public knowledge or any other means of investigation.

Consult to this end all enlightened men who might supply, confirm, validate or rectify the necessary information ([9], p. XIII).

The operation was a painful one. For the first time we hear the complaint of the industrial statistician, surprised by the difficulty of his discipline, which proves to be far greater than in the other domains, and full of bitterness towards the predecessors who neglected to put together good methods:

Some factories use only one sort of raw material from which ten different manufactured products are obtained, while others, on the contrary, make only a single product from ten raw or differently-produced materials. These anomalies place great difficulties in the way of the execution of statistical tables which, essentially constrained by analogy of type, symmetry in the grouping of figures and similarity of spacing, cannot lend themselves to these enormous disproportions. Nothing similar was found in agricultural statistics, the produce of the soil being easily encompassed by such expressions; nor was this disadvantage encountered in the old attempts at industrial statistics, considering that the obstacles were constantly kept at a distance by remaining on the surface of things ([9], p. XXI).

Reading between the lines, however, it seems that the statistician's heart was no longer in it - on the classifications side, of course. This publication reveals an insatiable interest in industry, a dazzled curiosity about its performances; the statistician constantly counts and recounts everything that can be done with a given quantity of labour and effort. Yet at the same time he is discouraged by the complexity of his subject, by the shiny new diversity of activities and products and the variety of their combinations, and so he more or less abandons any attempt at an aggregate level. The data on the large divisions (mineral, vegetable and animal products) look like afterthoughts, and there is no real attempt to avoid or even to indicate double countings of sales. Comparisons with previous censuses are rudimentary.

There are, on the other hand, veritable monographs on establishments. The following are listed for each establishment:

• the nature of the establishment, i.e. its activity; there is no indication of how it is determined;
• the municipality in which it is situated;
• the name of the manufacturer or producer;
• the rental value;
• the amount of business tax;
• the value of the raw materials used annually;
• the value of the products manufactured annually;
• the number of workers (men, women, children);
• average pay (men, women, children);
• power units (a picturesque list): mills (windmills, water mills, horse-driven mills), steam engines, horses and mules, oxen;
• burners: furnaces, forges, ovens;
• machines, whose curious breakdown illustrates the persistent dominance of the textile industry: looms, others, spindles.

The question of confidentiality never arises. Individual information is published without restriction. Moreover, if we accept that statistics consist of summarising a large number of individual information items in a single piece of information, then (apart from a few totals) the published results are not of a statistical nature.

This splintering of industrial statistics within the very framework provided by Tolosan can be explained by the historical situation. Industry is all-powerful and the pioneers have before them a vast field of activities about which they have few inhibitions at this stage. Competition between industrialists is not yet very lively; the State contents itself with indirect taxes: there is no point in concealing one's profits, and business people are proud to be making profits and showing them off. There is both a relative indifference to being known and a keen desire to know. The whole idea is to use the workforce and the equipment in such a way as to progress as far as possible; the novelty and the unknown factor are precisely the performance level which drives this entirely new industry, which is extending its sphere of activity to the detriment of the artisans it can so easily crush.

The performance is of far greater interest than the means. There is no need for aggregates.

Industrial statistics scatters its figures and headings throughout the "natural" framework constructed by Tolosan, and in so doing bursts its bounds. It renounces aggregates (i.e. the classification) in order to satisfy an industry in full expansion, eager to know every detail of its performances.

1861: the products

The 1861 industrial census marks a turning-point.

First, the classification used breaks completely with Tolosan; it employs "natural" groupings based above all on the purpose of products.

The first groupings by employer emerged in 1840. The struggle between the free-trade lobby and protectionists had led industries to form product family groups in order to defend their interests (8); attitudes had become accustomed to this viewpoint.

This time the census targeted "all factories or mills, irrespective of size" [10]. It had become impossible to publish results by establishment: instead, the district (arrondissement) was chosen as the geographical unit.

The large divisions of the classifications had to be abandoned because of the inextricable complexity (even then!) of integrating them (9).

The census classification describes an industrial world which is very different from our own. Although the most highly aggregated list of headings does not look too outlandish (10), the same cannot be said of their content.

For example:

The heading Wood industry (steam sawmills, water sawmills, veneer leaf, champagne corks, ordinary corks, barrels) has next to nothing in common with what we call mechanical woodworking today: the sawmill is nowadays allocated to agriculture and corks to "Various industries n.e.c.".

The heading Lighting: gasworks, tallow candles, resin candles, wax candles, tapers, corresponds to nothing we would expect today.

The heading Furniture (windows/mirrors, Gobelin carpets and tapestries, Neuilly tapestries, Meaux carpets, Aubusson carpets, Amiens carpets and broadloom, Utrecht velvet, Tourcoing carpets, felt carpets, wallpaper, waxed cloth, fluted chairs, cane chairs) has nothing in common - except the chairs - with what this word suggests today. The probable explanation is that dining-room and bedroom furniture etc. was made predominantly by craftspeople.

The same applies to the heading Clothing and toilet articles (strapped clogs, wooden clogs, slippers, shoes, hats [silk and felt], hats [felt and wool], felt hats, straw hats, caps, umbrellas, leather gloves [Millau], leather gloves [Grenoble], silk buttons, enamel buttons, buttons [mother-of-pearl and bone], combs and brushes, perfumery), distributed by today's classifications amongst perfumes, brush-making, the various clothing branches, hat-making, the footwear industry and glove-making - but includes nothing of what we regard as the "clothing industry" today.

It is worth quoting, out of interest, two headings which seem totally incongruous today:

Sciences and arts: white paper, white and grey paper, wrapping paper, cigarette paper, cardboard, pens, pencils, printed goods, rulers, spectacle frames, musical instruments, Jura watches and clocks, Montbeliard watches and clocks, Besançon watches.

Luxury and entertainment: decorations and imitation jewellery, cabinet work, playing cards, straw work.

This shows how dangerous it would be to try to compare one era with another without carefully scrutinising the homogeneity of classifications. This error is sometimes committed, however (as in [12], for example).

Despite all the upheavals, the approach is generally the same as it had been in 1841. Information enabling performance to be measured is sought in each district by detailed industry. This aspect is stressed in the introduction:

The second part is devoted to the study of the economic situation of the principal industries; its purpose is to determine, in the value of any manufactured product, the shares of the capital invested, the wages, the raw materials and the general costs comprising operating and administrative costs, taxes, insurance, and lastly profits.

In the third section we examine, according to the elements provided by the survey, the existing relationships between raw materials and manufactured products and, inter alia, the quantity and value of the work produced by a worker, a machine or a loom. This section constitutes a type of technological statistic of industry (1101, p- XV).

1895: the technologies

After the victory of the free-trade lobby in 1860, mistrust deepened between employers and the State. Refusal to respond became increasingly common; in parallel, liberalism became fashionable in government circles (11). The provision of information on industry was not enforced, and industrial statistics went into a long decline which lasted until 1940.

The middle of this period, 1895, did see an important advance in industrial statistics, however: Lucien March's processing and analysis of the population census. Since a proper industrial survey was impracticable, establishments were studied on the basis of census forms completed by the persons surveyed, on which they indicated their profession, the address of the establishment, etc. The principal activity of an establishment was determined by comparing the various forms referring to it. March made the best of a bad lot:

If one knows both the size of the labour force in an industry and the average annual wage of the workers in that industry, one can easily calculate the total wage bill. One then possesses a figure which may serve as an indicator to characterise the growth in value acquired by the raw materials in a particular industry without the need to ask heads of companies questions to which many would no doubt refuse to reply ([10], p. 11).

The problems inherent in determining the principal activity of the establishment from individual census forms (vague denomination, integration, juxtaposition) have since become traditional, but they are set out here for the first time.

Also for the first time we see the fleeting appearance of the association criterion: the attempt to construct aggregates in such a way that activities frequently associated within an enterprise fall within the same aggregate. Yet March absorbs this into what we call the technical criterion, according to which activities are grouped by the industrial technology used:

The industries or professions in each section are arranged in an order which separates as little as possible those which are most often grouped or juxtaposed in practice ... neighbouring industries in the classification are also analogous in terms of industrial processes. It is, in fact, the analogy of industrial processes that usually determines the association of several individuals in a single establishment or industrial centre. Even when a manufacturer wishes to offer his regular customers new articles whose application goes with that of the ordinary products he makes, he cannot do so without paying the closest attention to the working conditions of his workforce and machinery [10].

The classification compiled on that occasion remained in force, with a few alterations, for half a century. Its headings mix collective activities with individual ones:

The classification does not stop at the names of collective industries, but includes the names of special professions; for example, No 5460 covers couriers and delivery boys (with no other indication), but it should be noted that these are simply persons who, in the absence of adequate indications, could not be associated with the collective industry in which they cooperate; other persons of the same profession are classed under the number of the industry exercised in the establishment to which they belong.

Industrial statistics henceforth remained a by-product of censuses. A 1931 industrial survey attached to the census contained output results for the first time [11]. The classification came to be applied far less strictly than March had wished, and the mixture of individual and collective activities under the same headings made these less meaningful.

From 1940: revealing the decision-making centres

After the defeat in 1940, France tried to copy the German industrial organisation model; the Act of 16 August 1940 (12) turning its back on pre-war pseudo-liberalism and the abuse of individualism [7], [6], bestows important powers on the para-corporative and authoritarian Organisation Committees it creates. There was rapid amalgamation between the staff of the professional organisations, of the Organisation Committees, and of the raw materials distribution departments set up by the Act of 10 September 1940 ([6], pp. 84 and 85). Major work on industrial statistics (a by-product of various activities of the professional associations/Organisation Committees/distribution departments conglomerate) re-opened the question of the activities classification.

Collective and individual activities were separated, with individual activities placed in a separate classification known as the "occupational nomenclature".

Work began in 1942. Since the Organisation Committees existed and were both the "social partners" to the technologists and the sole intermediaries between State and industry, the classification begun in collaboration with the Committees reproduced the breakdown by employer. The Organisation Committees had been set up as quasi-cartels [6]; in general, employers grouped themselves according to an implicit application of the association criterion, which meant placing activities in the same enterprise under the same heading: the criterion was thus unspoken and went largely unrecognised.

This 1942 classification remained virtually unused. The work which would culminate in the 1949 nomenclature began in 1945. The two specialists responsible for compiling it were divided (13): one was already using the association criterion, while the other wanted to use the "purpose" criterion. A trace of this conflict is found in the preface to the 1949 nomenclature:

Three criteria, sometimes conflicting, inspired the classification of activities under the various headings: purpose, dominant raw material, technology. Although no hierarchy applied to these criteria, the "purpose" criterion generally prevailed. Thus, for example, the industries allied to the motor vehicle industry were classed with the latter; electricity, gas and water were juxtaposed as energy sources. However, since technology appears to take clear precedence over purpose in the real structure, the technology criterion was chosen; the manufacture of rubber footwear, for instance, was allocated to rubber production and not to footwear manufacture. Quarrying of construction materials was classed with quarrying of miscellaneous materials [20].

It was in fact the association criterion which, without ever being mentioned, decided whether, "in the real structure", "technology takes precedence over purpose".

Classification specialists do not live in an abstract world. Although a data-collection specialist like Mr Prevot would have defended the association criterion as a means of simplifying statistical surveys, this criterion also happens to meet the needs of economists remarkably well. The Liberation regime did not, in fact, bring a return to economic liberalism: far from it. The commissariat général du Plan was created in 1945-46. Initial attempts at economic synthesis then appeared in a national accounts context, and with a certain amount of planning in view. The planner's view of the economy became widespread (14). This raised the problem of the statistical unit, which was to be taken into account in compiling the information necessary to economic policy.

According to the depiction then devised, the relevant unit was the one able to react to an initiative from the center and hence possessing a certain autonomy: the enterprise.

It is clear that any attempt to describe and explain the trend in production must, if it is to be rational, feature the fundamental centers of decision-making: enterprises. A priori, any statistical system which breaks down enterprises into their technical components leaves out essential data, even if it is correctly tailored to its specific information objectives ([7], p. 35).

Branches and sectors

The study of classifications then expanded to assess the objects they cover. The shift in the focus of the statistician's interest sparked off the debates which took place on the contradiction between branches of activity (sets of parts of enterprises devoted to the same activity) and sectors (sets of enterprises).

The starting point was in fact a portrayal of the economy in which the "individual" (the "free" agent who "acts", "reacts", "behaves" etc.) is the enterprise: a sort of "decision-making atom". If we want to divide up the economy into sub-sets with a defined "behavior", such a portrayal obliges us to form sub-sets containing a whole number of enterprises: this then leads to a division of the economy into sectors. At the time, the need for this breakdown was felt almost physically. It was not thought then that enterprises could suitably be chopped up, or that they could be too big, or that the legal boundaries did not mean much, any more than it would occur to us nowadays to split a population of human beings into sub-sets which did not contain a whole number of individuals.

The concept of the branch did exist, however, and most of the econometric models (particularly those of Leontieff) were actually conceived within its framework. Herein lies the contradiction: on the one hand a portrayal of the economy which obliges the user to consider sectors (sets of enterprises) as agents useful to economic analysis; on the other hand an economic theory which provides appropriate models only for the study of branches, statistical entities which have no representational meaning.

A controversy thus emerged from 1951 to 1955 between those in favor of a depiction founded on a sectoral breakdown and those who wanted a branch breakdown. Analysis of the technologies used in industry naturally fits the branch concept, as do Leontieff's models. An analysis of decisions, investment etc. falls more naturally into the sectoral framework, however. Models inspired by Mr Guilbaud tended to isolate decision-making centers and so referred to sectors. Lastly, discussions at the time also reflected the diverse behavior of heads of companies with regard to data collection on enterprises. Some preferred the branch framework, which was less transparent, while others preferred the sectoral framework, which was useful for studying the problems posed by mergers. The branch proponents included those responsible for industrial statistics and for the Plan; the sector advocates included Mr Gruson and his "team", the Finance department and some economics researchers.

The economic table compiled in 1951 [13] following these debates was a complex one using both sectors and branches, defined with reference to product groups. An initial table showed sectoral sales by product group; another showed purchases. The procedure for drawing up these tables was so cumbersome that the sector/product cross was abandoned in 1956 [14] and replaced by the branch framework.

The advantage of the association criterion

As can be demonstrated [17], the use of the "association criterion" minimises the difference between branch and sector statistics by mitigating the contradiction between sector and branch. It was thus particularly well suited to the requirements of planning, which identified enterprises as economic agents while using the resources of branch-based econometrics.

When the nomenclature was revised in 1959, it was explicitly mentioned that:

We attempted in this project to throw more light on the branches of economic activity in which establishments or enterprises are active, in such a way that each heading covers comparable establishments and constitutes a meaningful economic unit ([21], pp. 7 and 8).

This is still a bit vague. It was not until 1962 that Mr Prevot gave the association criterion its definitive formulation:

The closer the sets of elementary activities defining branches are to the groupings of elementary activities actually found within industrial establishments, the fewer the chances of error. The identity of the activity groupings defining the headings of an industries classification with the most frequent activity groupings in industrial enterprises or establishments may therefore be deemed an essential precondition to the compilation of an industries classification [16].

3 - The current picture

We have already described the situations of the past. An in-depth look at the current landscape will illustrate its ambiguities and complexity. It would be wrong to assume that all specialists advocate and accept the association criterion. In reality it corresponds only to data-collection logistics and to the needs of a planned economy in a portrayal which identifies enterprises as independent agents.

The depictions which in the past successively favored the "raw materials", "technology" or "product" viewpoints are still very useful for studying other problems; moreover, the use of classifications for regulatory or administrative purposes means adopting other constraints.

A single compromise

One solution, of course, would be to use several different classifications, each tailored to a particular purpose, but all linked via a common level. The chosen route, however, was to arrive at a single compromise by arbitration between the various requirements.

• The least common viewpoint these days advocates a classification founded on raw materials or products consumed: everything made from the same input would be grouped under the same heading. This criterion rests on the concept of the economy as a technique for using rare resources and on the memory of the shortages of the Second World War, when supply problems had to be given top priority. It is well suited to the examination of these problems.
• Far more people prefer to group activities by the technology used. This approach sees things from the material and human investment angle; engineers frequently take this approach. It fits the examination of equipment problems particularly well.
• Economists often favour the product criterion, which places under the same heading activities requiring very varied technologies and raw materials but supplying analogous products. The "product" viewpoint rests on a portrayal of the economy in which the market is the dominant element. It agrees with the marginalist economic doctrine currently in vogue ("demand drives supply"), and also with the idea, widespread nowadays, that market problems are becoming serious.
• The use of the association criterion, as pointed out in the preface to the 1949 classification [20], leads to the alternate use of the above criteria. They come together as specific cases: when the most difficult task for enterprises really is to create a market, they will tend to specialise in a product and will use all possible technologies and raw materials to manufacture it. The use of the association criterion then means grouping under the same heading the elementary activities contributing to the manufacture of a single product with the help of various technologies: the priority will thus be given in this particular case to the "product" criterion. Hence the French activities nomenclature constructs the heading "games and toys".

Similarly, when the technology used requires heavy investment or highly specialised staff, or both, industrialists will tend to have it produce the entire range of possible products: if they are seeking to expand, exploring new markets is easier than acquiring and mastering new equipment. The association criterion would then mean grouping the manufacture of various products by the same technology and so giving priority to the "technology" criterion. The French activities nomenclature constructed the "Processing of plastic materials" heading for this purpose.

It also happens - seldom nowadays, but that would change if there were a shortage - that the "raw materials" criterion is favored by an analogous mechanism. Hence, the French activities nomenclature contains the heading "Rubber products industry".

To summarize, the association criterion leads to the partial use of the product, technology or raw materials criteria when these are crucial to the activities concerned.

Institutional constraints

A combination of a variety of criteria - technology, material, product, and association as a compromise between the first three - requires a "preclassification". To become the official classification, this needs to be amended and compared with other "preclassifications". There are two reasons for this: technical and institutional. Let us take a brief look at the technical reasons, followed by a longer look at the institutional ones:

• Classifications are expected nowadays to last for some time (usually ten years). Clearly a classification must fit not the industry of today, but the industry which will prevail in five years' time. We therefore have to make assumptions about the trend in industrial structures. Moreover, the pointers given by the association criterion have to be interpreted from the point of view of their statistical meaning. Lastly, there must be good consistency with foreign classifications. This obligation, though technical, is already halfway to the institutional viewpoint.
• Those responsible for classifications tend to adhere increasingly to the institutional view, claiming that the classifications studied have more than just statistical or informative aims, since they also serve management and legislation, and compiling a classification does not mean constructing a theoretical tool for an abstract definition of reality. The classification is essentially a tool for management, for data collection and processing, and for study. A complex task is thus defined. The classification must be a management tool: what we need is not an instrument of pure ("abstract") knowledge, but something that gives managers - "decision-makers" - the easiest means of collecting and analysing data.

The classification is not directly a decision tool; it is directly a study tool, even though the goal of the study is not pure knowledge, but to meet the needs of the decision-makers.

For the decision-makers to need pure knowledge - a finely-tuned, purely theoretical instrument - would be no contradiction, but there are several reasons why this does not happen. The main one is doubtless that theoretical knowledge progresses very slowly and is not easily constrained to meet predetermined deadlines, whereas the need for action is always urgent. Action-takers thus see institutions as the quickest and most easily available framework, even if its contours are questionable.

Let us take as an example something that really exists but shall remain nameless. Let us consider a branch of industrial activity that resembles a set of activities which were rightly juxtaposed a long time ago, but have since seen such technical transformations that, from the point of view of pure economics, the aggregate ought to be broken down, most activities placed under other headings, and only a small core kept under the original denomination.

The economist, armed as it happens with an association criterion, is ready to make a high-handed cut, but the decision-makers do not like it. The branch which has disappeared from the economic landscape is still there as an institution. The industrialists, linked by a sort of family bond, continue to meet under the banner of a professional organization which is both their club and their representative in dealings with the State, the trade unions, other associations affiliated to the CNPF, and with foreign colleagues and competitors; this organization has a professional journal which is read, which influences the profession, etc.

All of this does not rule out internal conflict or the slow decay induced precisely by economic change, but the branch still exists as an institution, as a spokesman for its decision-makers; and ultimately, this aggregate which is of no interest to the pure economist may be a valued and noteworthy partner for a planner or the head of a large company.

Moreover, the institution itself would not easily tolerate being erased as an economic aggregate, and is good at voicing its opposition.

Another example: the classification serves to define the scope of certain regulations, certain subscriptions; it is used by government to compile records which may be extremely voluminous. Changing it may provoke earthquakes, and therefore change is avoided as far as possible. This institutional constraint is less obvious than the former, but it exists nevertheless.

*     *

Thus we see that classifications are at the heart of the contradiction between scientific and administrative activity. The latter may take scientific form, since it uses the tools provided by science, but its aim is short-term and medium-term action that cannot tolerate the slow advance of science.

Classifications pose, and also solve in their own way, a problem of compromise between two contradictory points of view, each of which is valid.

These two points of view are no less interdependent for being contradictory, however. As the first section attempted to show, the framework of the portrayal serving statistical practice cannot avoid being closely linked to the very general findings which govern such portrayals, and hence, in particular, to those of economics. The problems posed by this interdependence appear as soon as we get past the purely formal aspect of the "classifications" object and start thinking about its meaning. The second part sought to demonstrate how this interdependence showed itself historically, but also how it came to be mediatised by what is commonly known as administrative inertia and red tape. A compromise on the tools of statistical practice, here classifications, was always reached between the requirements of current theory and the practical and institutional means available to statistical agents or those they interview: hence the ultimate ambiguity and difficulty not only of classifications, but also of the collection and presentation of statistical results in the framework they define.

Moving parts

In industrial statistics, the statistical individual itself is hard to pin down. A human being can be identified unambiguously by her civil registration number, including her place and date of birth and serial number, but a business is built on shifting sand: its location, activity and name may vary. It is only ever a place through which individuals, materials and money flow, but this place is as volatile as an eddy in a stream.

Nothing is stable. No matter how we split up the industrial economy - by unit size, region, activity - it will always be made up of moving parts. Constructing sampling frames for enterprises is a shot in the dark: the concepts used are vague and always questionable. We saw an example of this with classifications.

If we re-read the publications of industrial statisticians of the past, or even those of today, we are struck by the disproportion between the difficulty of the object and the paucity of the means employed to study it, whether the object is professional organizations or government. Yet we never wanted to recognize the difficulty. We believed that counting was enough; basically we assumed that a goods item was a simple, obvious object - which it is not! - and that simple methods would enable us to cope with its production.

We often find both surprise and consternation flowing from the pen of the industrial statistician about the difficulty of his object and the fuzzy indications it yields. It seems that he himself set out in all confidence and found himself confronted by a monster of complexity.

It is thus an illusion that has to be abandoned. Industrial statistics is a difficult domain. Although the pioneers that have explored it for the past thirty years have done the bulk of the work, we have not yet emerged from their era.

We must not therefore be surprised if, as Desrousseaux [8] says, the most sensible discussions of the industrial economy are often those of literary economists. In the absence of a really sophisticated method of portraying industrial reality, the statistician has to rely on tools which are still far too rudimentary.

Progress under way

It is to be hoped that the current methodological progress will encourage a dialectic between the quality of data and the adequacy of their interpretation. In all, grasping the imperfection of statistical methods is a measure of the possibility of correcting them. And if one symptom of the scientific character of a methodology is the cumulative nature of its progress, industrial statistics methods would seem promising. This is why, despite all the difficulties discussed theoretically in the first section and historically in the second, the sheer imperfection of quantifying certain industrial phenomena has an air of scientific validity, provisional as are all scientific conclusions.

There is a striking contrast between the high-quality and refined use of data and the scant attention given to the molds in which these data are set. But the solidity of the product depends on the quality of the mold. The very plan of the industrial statistics edifice is formed by this mold: the classifications. That edifice, like industry itself, is a site of constant and unpredictable change.

The theoretical and administrative problems of the years to come will be arduous. Statistical investigation needs finer, more precise and higher-quality tools. The emerging discussions on this subject testify to the existence of these needs, whose development portends systematised effort which will compensate for the qualitative imbalance between considerations of form and of content.

Bernard Guibert belongs to INSEE's "Economic records and classifications" subdivision. Jean Laganier and Michel Volle have been placed by INSEE at the disposal of the French Ministry of Industrial and Scientific Development (Statistical Office of the Miscellaneous Industries and Textiles Directorate).

Bibliography

Philosophy of knowledge

[1] Pierre Bourdieu: Éléments pour une théorie sociologique de la perception artistique
[2] Michel Foucault: Les mots et les choses
[3] Henri Lefebvre: Logique formelle, logique dialectique
[4] Gaston Bachelard: Le rationalisme appliqué
Economic history and economics

[5] Roger Priouret Les origines du patronat français
[6] H.W. Ehrmann La politique du patronat français (1936-1955)
[7] Claude Gruson Origines et espoirs de la planification française, Dunod 1968
[8] Desrousseaux L'évolution économique et le comportement industriel
Statistical results

[9] Statistique générale de la France "Industrie 1847"
[10] Statistique générale de la France "Industrie, enquête de 1861-1865 ", Paris, 1873
[11] Statistique générale de la France "Recensement de 1896. Industrie"
[12] Statistique générale de la France "Enquêtes annexes du recensement de 1931", Paris, 1935
[13] SEEF "Les comptes de la nation pour l'année 1951"
[14] SEEF " Les comptes de la nation pour l'année 1956"
[15] INSEE, Tendances de la conjoncture, No 8, 1970
Articles on classifications

[16] J. Prévot: "Réflexions sur les problèmes des nomenclatures statistiques d'industries et de produits", Informations statistiques Nos 1 and 2, 1962.
[17] Benier, Giudicelli, Grosmangin, Guibert, Laganier, Masson, Pietri, Rousseau, Volle: "Une construction des objets de l'analyse économique, les nomenclatures de l'industrie", paper drawn up for the French Ministry of Industrial and Scientific Development
[18] Benier, Giudicelli, Grosmangin, Guibert, Laganier, Masson, Pietri, Rousseau, Volle: "L'analyse des données et la construction des nomenclatures d'activités économiques", Annales de l'INSEE, No 4.
Classifications

[19] Nomenclature douanière (French Customs Nomenclature)
[20] Nomenclature des activités économiques (French Economic Activities Nomenclature) (INSEE, 1949)
[21] Nomenclature des activités économiques (French Economic Activities Nomenclature) (INSEE 1959)

Footnotes (1) "Essay: A work in prose, very freely structured, dealing with a subject which it does not treat exhaustively". Le petit Robert.

(2) A division of a set E is a family of sub-sets such that the union of all the sub-sets forms set E and the intersection of two separate sub-sets is empty. A division is said to be embedded in another if any sub-set of the first is included in one of the sub-sets of the second. An order relationship (higher or lower levels) can be determined between the divisions which determine a classification [17].

(3) For the examples we have quoted, the code numbers used in the customs nomenclature are:
29. Organic chemicals.
29-38. Natural or synthetic provitamins and vitamins, etc.
29-38-23. B2 vitamins.
61. Clothing and clothing accessories in fabric.
61-02. Women's, girls' and young children's outerwear.
61-02-49. Cotton coats for women and girls.
73. Cast iron, iron, steel.
73-32. Bolts and screws etc. and similar articles in cast iron, iron or steel, etc.
73-32-21. Machine screws for wood, threaded, in bulk, shank diameter not more than 6 mm.

(4) Depending on the domain, an individual may be a human being, an animal, an activity, a commodity etc. For some authors, only this list of elementary headings constitutes the classification, to the exclusion of any idea of groupings. We will not make this view our own.

(5) Two questions then arise: what is the theoretical significance of the classes thus distinguished, and what is the the relevance of their particular structure and that of the field as a whole? What, for example, does the thought-object structure of an autonomous economic field mean from a theoretical point of view? Is it relevant in terms of a theory embracing the whole socio-economic field?

(6) Man-made and synthetic textiles are, in fact, classed as chemicals in the German nomenclature.

(7) The Statistique de la France had been reinstated by Bonaparte in 1800. The published results were systematically calculated at the level of the Empire as a whole, of which France proper was only a part. Chaptal therefore had to make estimates in order to assess the size of French industry.

(8) A single industrialist may very well be both a protectionist of his/her products and a free-trader in terms of raw materials and production factors: this was then the case with textile manufacturers [5].

(9) "The administration had very much hoped to be able to draw up a detailed classification by type of industry. Yet on many points and chiefly in the large manufacturing establishments, since the various operations involved in the full execution of an industrial product were found together in the same factory, each branch of industry had not always been designated. For textiles, for example, we had often lumped together, under the same heading, combing, spinning, weaving, finishing, etc.; and for metallurgy blast furnaces, forges, foundries and so forth. We therefore recognised the impossibility of drawing up a detailed classification, and had to be content with forming general groups, taking care to make a clear separation between dissimilar industries" [10].

(10)

1. Textile industry.
2. Extractive industry.
3. Metallurgy.
4. Metal objects.
5. Leather industry.
6. Wood industry.
7. Ceramics.
8. Chemical products.
9. Building.
10. Lighting.
11. Furnishings.
12. Clothing and toilet articles.
13. Food.
14. Means of transport.
15. Sciences and arts.
16. Luxury and entertainment.

(11) " The ideas of a high-ranking official of the time ... were strictly liberal. The idea that the State could usefully intervene in economic life was forcefully rejected." [7].

(12) Known as the "Bichelonne Law".

(13) Mr Prevot and Mr Duon.

(14) Note that "planner" does not necessarily mean "statist". Professional organisations and large companies had - and still have - exactly the same representation.