The previous plate drew the seller's blade as a single smooth curve and let the matter rest there. But the cost that the curve summarises is not one simple thing, and the careful use of the margin turns on keeping its kinds apart. Two confusions in particular attend the reading of a cost, and each is among the most common and most expensive errors in all of practical reasoning — so common that the discipline of avoiding them is, for most people most of the time, the single most useful thing economics has to teach. The first is to mistake the cost of the average unit for the cost of the next one. The second is to let a cost that has already been paid, and cannot be unpaid, weigh upon a decision about what to do now. Both are failures of the same kind, the kind the margin exists to prevent; and this plate, which examines them, is where the margin stops being a theory of how prices are formed and becomes a rule for how choices ought to be made.
Take the first confusion. The marginal cost of a thing is what it costs to produce one more of it — the extra fuel, the extra hour, the extra ton of material that the additional unit requires, and nothing else. The average cost is the total cost of all the units divided by their number, the cost per unit taken across the whole run. These are different quantities, and for the purpose of deciding whether to make one more they can differ to the point of opposite advice, because the average is weighed down by costs that the marginal unit does not bear. Consider an aircraft about to depart with empty seats. The average cost of a seat on that flight — the fuel, the crew, the aircraft, the landing fees, all divided by the number of passengers — might come to two or three hundred pounds. But the marginal cost of carrying one more passenger in a seat that would otherwise fly empty is almost nothing: a little more fuel for the added weight, a meal, a few grams of paperwork, perhaps twenty pounds in all. An airline that refused to sell that last seat for fifty pounds, on the ground that fifty is far below the two hundred a seat costs, would be making a plain and costly mistake, for the fifty exceeds the twenty that the seat actually costs to fill, and the flight is leaving regardless. The average said no; the margin said yes; and the margin was right.
The reason the average misleads is that it answers a question the decision never asked. The decision asks: what will it cost me to do this one further thing, and is that less than what the thing will bring? It is a question entirely about the increment, about the difference between acting and not acting, and the only costs that bear upon it are the costs that the increment itself adds. The average, by contrast, reports the cost of a typical unit across the whole of past and present production, and into that figure are folded all the fixed costs of being in business at all — the aircraft, the factory, the standing overhead — costs that were going to be incurred whether or not this last unit is made, and that the last unit therefore does nothing to change. To charge the marginal decision with a share of those fixed costs, by judging it against the average, is to burden it with expenses it does not cause. The average has its uses; it tells you, over the long haul, whether the whole enterprise is covering what it costs to run, which is a real and important question. But it is not the question a marginal decision asks, and to answer the one with the other is to invite exactly the error of the empty seat.
The average cost answers a question no decision ever asks — what the typical unit cost — when the only question a decision asks is what the next one will cost; and the two answers can point in opposite directions.
The second confusion runs deeper, because it is rooted not in an accounting mistake but in something close to a moral intuition, and it concerns costs that lie not ahead of the decision but behind it. A sunk cost is a cost already incurred and beyond recovery — money spent that cannot be unspent, effort expended that cannot be returned, whatever one chooses to do from this moment forward. The principle that governs such costs is as simple to state as it is hard to obey: a sunk cost is irrelevant to any decision about the future, and ought to be wholly ignored. What is gone is gone in equal measure whether one presses on or turns back; it therefore cannot favour either course, and the only costs and benefits that can rightly weigh upon the choice are those that still lie ahead and still depend on what one does. The classic monument to the violation of this principle is the supersonic airliner Concorde, which Britain and France went on funding long after it was certain the project would never repay its costs, on the express ground that so much had already been spent that to stop would be to waste it — as though the money already poured in could somehow be retrieved by pouring in more. The half-finished, hopeless venture pursued because of what it has already consumed is the sunk cost fallacy, and it has a name in every language because it is committed in every walk of life: the bad book finished because it was bought, the losing investment held because selling would make the loss real, the war prolonged so that those already fallen shall not have died in vain.
That the fallacy is so easy to name and so hard to resist tells us that something in the mind is built to commit it, and the behavioural economists who have studied the matter have found several somethings. There is loss aversion, the well-documented asymmetry by which a loss looms larger than an equivalent gain, so that the act of abandoning a project — which forces one to record the loss as final — is felt as more painful than the continued, deferred bleeding of pressing on. There is the wish to appear consistent, to oneself above all, since to abandon a course is to confess that embarking on it was a mistake. There is the mental accounting by which we keep a separate ledger for each undertaking and recoil from closing any ledger in the red. And beneath them all there is the simple, stubborn feeling that to walk away is to waste what has been spent — a feeling that inverts the truth, for the spending is already wasted, and walking away wastes nothing further, while pressing on in order to honour it wastes more. The pull of the sunk cost is thus not a failure of arithmetic but a quirk of the heart, and the marginal discipline that overcomes it — attend only to what lies ahead — is correspondingly difficult, a thing one must practise against the grain of one's own instincts.
The money already spent is the one consideration that no decision now can recover, and is therefore the one consideration a sound decision must ignore — yet it is precisely the one the mind clings to, calling it a reason to go on.
It is worth seeing that the two confusions, though they wear different faces, are at bottom the same error, and admit the same cure. The trap of the average and the trap of the sunk cost are both the trap of letting a cost that does not change with the decision masquerade as a cost that does. In the average, the intruder is the fixed cost, spread across the units and thereby smuggled into the reckoning of a unit it does not touch. In the sunk cost, the intruder is the past expenditure, gone whatever happens and thereby irrelevant to a choice it can no longer affect. Both are costs that the decision is powerless to alter, and a cost the decision cannot alter is a cost the decision must exclude, for the whole of rational choice consists in comparing the things that change between one course and another, and a thing that stands the same on both sides of the comparison drops out of it entirely. The margin, properly understood, is precisely the name for what changes — the difference an action makes — and to think at the margin is to attend to that difference alone, neither inflating it with fixed costs it does not cause nor anchoring it to sunk costs it cannot recover. The two errors are one error, and the margin is its one correction.
The rule wants two honest qualifications, lest it be pressed further than it will go. The first is that costs are not always as sunk, or as fixed, as they look: a payment may be partly recoverable, an asset partly resaleable, and to the extent that some of what was spent can still be got back, that part is not sunk and does bear upon the choice. The discipline is to ignore what is truly gone, not to wave away what might yet be salvaged. The second is subtler, and concerns commitment. There are situations in which the visible willingness to honour a sunk cost — to fight on past the point of profit — has a forward-looking value of its own, because it builds a reputation for resolve that deters future challenges; a state that abandons every costly venture the instant it turns unprofitable may invite others to test it. But this is not, on inspection, an exception to the rule so much as a confirmation of it, for what justifies pressing on in such a case is not the past cost at all but the future benefit of the reputation, a forward-looking gain weighed in the proper marginal way. And the average, too, keeps its proper domain: in the long run, when the fixed costs themselves become variable and the firm must decide whether to exist at all, it is precisely whether revenue covers average cost that settles the question. The margin governs the decision at hand; the average governs whether there should be an enterprise to make it.
With this plate the margin completes its descent from the heights of value theory to the floor of daily choice. It began as the key to a paradox and the engine of a revolution; it became, in Marshall's hands, the apparatus of supply and demand; and it arrives here as a plain rule for anyone who must decide anything — weigh the cost of one more against the gain of one more, count only what the decision itself can change, and let the average and the sunk alike fall silent. But there is one further place the marginal reckoning of cost must be carried, and it is the most consequential of all. For a firm deciding how much of a thing to use — how many workers to hire, how many acres to rent, how much capital to employ — the cost of one more unit of the factor must be set against what that one more unit adds to its output, the factor's marginal product; and the doctrine that each factor is worth, and tends to be paid, exactly the value of what one more unit of it produces leads directly back to the first family of this table, and to the oldest of its prices. The marginal product, and the road it opens back to the wage, is the subject of the next plate. As the table says of every quantity, first to last: the argument remains open.