Total versus per-unit: the same cost wearing two faces
In the last two guides you learned to tag every cost against a cost object and to sort it into one of three kinds — fixed, variable, or mixed. That was classification: putting each cost in a box. Cost behavior is the moving picture — it asks what each box *does* as the level of activity (units made, hours run, meals served) climbs and falls. And the single fact that trips up almost every beginner is this: a cost behaves one way in *total* and the opposite way *per unit*. You have to hold both faces in mind at once.
Take a variable cost — the leather in a workshop's handbags, say, at 20 dollars a bag. Make one bag and total leather cost is 20 dollars; make a hundred and it is 2,000. In total it rises in a straight line with output. But per bag it never moves: it is 20 dollars whether you make one bag or a thousand. That is the signature of a variable cost — *total changes, per-unit stays flat*. Now take a fixed cost — the 6,000-dollar monthly rent on the workshop. In total it does not budge whether you make ten bags or ten thousand. But per bag it melts away: 600 dollars per bag at ten bags, 6 dollars per bag at a thousand, less and less as you spread it over more units. *Total stays flat, per-unit changes.* Fixed and variable are exact mirror images, and the mirror is the total-versus-per-unit divide.
Lay the two side by side and the mirror is unmistakable. As volume rises, the *total* variable cost climbs in a straight line while its *per-unit* figure stays pinned at 20 dollars; the *total* fixed cost holds dead flat at 6,000 dollars while its *per-unit* figure slides downward — 60 dollars a bag at 100 bags, just 10 dollars a bag at 600 bags, all from the very same 6,000-dollar rent. Each cost type holds one face still and lets the other move, and which face is which simply swaps between the two.
The relevant range: the honest fence around your assumptions
Everything above quietly assumed two things: that the rent is fixed *no matter what*, and that the leather costs exactly 20 dollars a bag *no matter how many* you make. Push hard enough and both assumptions break. Make ten times as many bags and you outgrow the workshop — you must rent a second floor, and suddenly the "fixed" rent jumps to a new, higher flat level. Buy leather by the truckload and the supplier gives you a volume discount — the "constant" per-bag cost drops. Real costs are only fixed or neatly variable across a *band* of activity, not across all of infinity.
That band has a name: the [[relevant-range|relevant range]] — the span of activity over which your cost-behavior assumptions actually hold. Inside it, fixed costs really do stay flat and variable costs really do stay linear, so the simple straight-line picture is trustworthy. Step outside it and the lines bend, kink, or jump to a new level, and any number you calculated from the old assumptions becomes a guess dressed up as arithmetic. The relevant range is not a flaw in the model; it is the model honestly stating where it is allowed to speak.
A vivid case is the step cost — a cost that is flat for a while, then leaps to a new flat level. One supervisor can oversee up to 50 workers; the 51st worker forces you to hire a second supervisor, and supervisory salary jumps in a single step. Within one tread of that staircase, the cost behaves as a textbook fixed cost; across the whole staircase it is anything but. The relevant range, in practice, is usually one tread wide — the slice of volume your firm is realistically going to operate in over the planning horizon. Choose that slice honestly and the simple model serves you well; pretend it stretches forever and it will quietly lie to you.
Mixed costs, and prying them apart
Most real-world costs are not purely fixed or purely variable — they are mixed, carrying both a fixed lump and a variable rate inside one bill. A delivery van costs a flat amount each month for insurance and registration whether it leaves the garage or not, plus fuel that rises with every kilometre driven. A phone plan has a fixed monthly line charge plus a per-minute usage fee. Written as a formula, every mixed cost is the same shape: a starting amount that exists at zero activity, plus a slope that adds cost for each extra unit of activity.
Before you can forecast a cost, plan a budget, or ask the cost-volume-profit questions waiting in the next guides, you have to split a mixed cost into its fixed lump and its variable rate. The quickest hand tool for the job is the [[high-low-method|high-low method]]. The idea is disarmingly simple: take only two data points from your records — the period of *highest* activity and the period of *lowest* — and let the difference between them reveal the variable rate. Because the fixed portion is the same in both periods, it cancels out in the subtraction, leaving only the variable cost of the extra activity. Divide that change in cost by the change in activity and you have the cost per unit of activity — the slope.
- Find the highest-activity period and the lowest-activity period in your data (by activity level, not by cost).
- Variable rate = (cost at high − cost at low) ÷ (activity at high − activity at low). The fixed part cancels in the top and bottom differences.
- Fixed cost = total cost at either point − (variable rate × that point's activity). Both points must give the same fixed figure.
- Write the cost equation: Total cost = fixed cost + (variable rate × activity), good only inside the relevant range you observed.
Electricity bill vs. machine-hours over six months:
Highest month: 800 machine-hrs -> $5,000
Lowest month: 300 machine-hrs -> $2,500
Variable rate = (5,000 - 2,500) / (800 - 300)
= 2,500 / 500 = $5 per machine-hour
Fixed (use high pt) = 5,000 - (5 x 800) = 5,000 - 4,000 = $1,000
Check (use low pt) = 2,500 - (5 x 300) = 2,500 - 1,500 = $1,000 OK
Cost equation: Total = $1,000 + $5 x (machine-hours)Why two points are not enough: the scatter plot
Be honest about what the high-low method just did: it threw away every data point except two, and bet the entire cost equation on those two. That is its charm — you can do it on the back of an envelope — and also its danger. If either the highest or the lowest month happened to be a freak (a heatwave spiked the electricity, a flood shut the line), the whole line tilts off the truth, because those two outliers are doing *all* the work. High-low does not look at the other four months at all; it cannot tell a typical point from a fluke.
The cheap, powerful safeguard is a scatter plot: put activity on the horizontal axis and cost on the vertical, and drop a dot for every period. In a few seconds your eye does what no formula can. If the dots fall roughly along a straight line, the cost really is behaving in a linear, mixed way and the high-low equation is believable. If the high or low point sits far off the cloud of the others, you have just spotted the outlier that would have poisoned high-low — and you can exclude it. If the dots curve, or split into two separate clusters, you have learned that a single straight line is the wrong model entirely, perhaps because you have drifted across the edge of one relevant range into the next.
Why all this work is worth it
Splitting costs into fixed and variable is not academic busywork — it is the foundation the rest of managerial accounting is built on. The moment you know a product's *variable* cost per unit, you can subtract it from the selling price to get the contribution margin you met earlier — the dollars each sale throws off to cover the fixed lump and, beyond that, to become profit. And once you know the total fixed cost and the contribution margin, you can answer the questions that the next guides are entirely about: how many units must we sell just to stop losing money, and what happens to profit when volume changes?
That family of questions is [[cost-volume-profit-analysis|cost-volume-profit analysis]], and the volume at which total contribution margin exactly covers total fixed cost — where profit is zero — is the [[break-even-point|break-even point]]. None of it works without the behaviour split you have just learned, and none of it works outside the relevant range that fences it in. If a CVP calculation tells you to sell 40,000 units when your factory tops out at 25,000, the answer is not a target — it is a sign you have stepped beyond the range where your fixed costs and your linear variable costs still hold, and the whole model has quietly stopped being true.
So carry three things forward. First, every cost wears two faces — flat one way, sloped the other — and you must always say which one you mean. Second, fixed, variable, and mixed behaviour are simplifications that earn their keep only inside the relevant range; state the range or the number is bluffing. Third, the high-low method gives you a fast split and a scatter plot keeps it honest — and together they hand you the variable rate and fixed lump that every decision tool downstream will demand.