General knowledge — expertise, statistics, etc. — is usually more economically used by the higher decision-making units. For example, a decision-making unit which encompasses five subordinate units can acquire a given expertise and statistical data which it applies to all five units, whereas if each unit had independently acquired the same expertise or statistical data it would have cost five times as much altogether. For this kind of knowledge, the cost advantage tends to be on the side of the larger and higher decisionmaking unit. But for highly specific knowledge — the local life style, the reliability of particular suppliers, the level of skill of a given executive, etc. — the subordinate units immediately in daily contact with the relevant facts can much more easily and more cheaply synthesize the knowledge and draw inferences.

It is unnecessary to attempt any general rule as to where the overall balance lies in comparing the respective costs of knowledge in larger and smaller decision-making units. What is important is to understand that (1) the respective cost advantages of the large and small units differ according to the kind of knowledge involved (general versus specific), that (2) most decisions involve mixtures of the two kinds of knowledge, so that the net advantages of the larger and smaller units vary with the kind of decision, and (3) the effectiveness of hierarchical subordination varies with the extent to which the subordinate unit has knowledge advantages over the higher unit. In those cases where the subordinate unit has better information, then in terms of the whole decision-making process the knowledge is one place and the power is another; the quality of decisions suffers as a result. Moreover, subordination itself becomes illusory to the extent that the lower level unit can use its knowledge advantages to evade, counteract, or redirect the thrust of orders from its nominal superiors.

Some examples from various institutions and various societies may illustrate these crucial points. Agriculture has its general principles and statistics, but agricultural production involves much highly specific knowledge about the characteristics and contours of particular plots of land, and about the freshness, flavor and keeping qualities of specific batches of fruits, vegetables and dairy products — all of which are changing by the hour. No expert can say from 100 miles away, and sight unseen, that this year’s grape crop is good, or even that last week’s good grapes are still good this week. By contrast, an expert on the manufacture of steel can specify the exact quality of steel that will be produced by given combinations of iron ore and coal at given temperatures. For these reasons, steel production has been successfully centrally planned and controlled in various countries, whereas agricultural production has had such chronic problems and periodic disasters in centrally planned economic systems that even the most centralized communist governments have had to make major exceptions in agriculture, allowing decentralized decision-making of various sorts.

For similar reasons, in capitalist countries it is common to have chains of grocery and department stores selling standardized items, but there are no large chains of high quality restaurants of a sort which depend upon atmosphere and finely prepared food. Such restaurants require constant attention to the demeanor of the staff and the delicacy of the chef, and those cannot be effectively controlled by distant experts. Usually the owner and manager of a successful restaurant of this sort is right on the premises, often from the moment it opens each day to the moment it closes at night. By contrast, the top executives of Sears or Safeway cannot and need not be present at their hundreds of stores across the country, for much of the knowledge they need can be gained from statistics, experts, and accounting data.

THE STRUCTURE OF INCENTIVES