You're staring at a spreadsheet. Column B has total costs. Now, column A has units produced. You plot them, and — surprise — the dots form a pretty straight line It's one of those things that adds up. Practical, not theoretical..
Textbook stuff. Clean. Predictable.
But here's the thing: that line is lying to you. Or at least, it's only telling the truth under very specific conditions That's the part that actually makes a difference..
What Is Linear Cost Behavior
Linear cost behavior means total costs change at a constant rate per unit of activity. The slope is your variable cost per unit. Graph it, and you get a straight line. The y-intercept is your total fixed costs.
Simple equation: y = a + bx
Where y is total cost, a is fixed cost, b is variable cost per unit, and x is activity level.
But — and this is the part that gets skipped in intro accounting — that equation only holds inside what accountants call the relevant range Small thing, real impact. Still holds up..
The relevant range is everything
Most textbooks define it in one sentence and move on. In practice? It's the whole ballgame.
The relevant range is the band of activity where your cost assumptions actually hold. Fixed costs stay fixed. That said, variable costs per unit stay constant. But step costs don't step. Here's the thing — economies of scale don't kick in. Overtime premiums don't appear Small thing, real impact..
Step outside that range, and linearity breaks.
Produce 10% more widgets? Produce 300% more? Now you're adding a second shift, negotiating new supplier contracts, maybe leasing another building. Your "fixed" costs just jumped. Sure, variable costs scale nicely. Your "variable" cost per unit might drop from volume discounts — or spike from overtime Easy to understand, harder to ignore..
The line bends Not complicated — just consistent..
Fixed costs aren't fixed forever
This trips people up constantly. "Fixed" doesn't mean "never changes." It means "doesn't change with volume inside the relevant range.
Rent is fixed — until your lease renews at double the rate. Insurance is fixed — until you add a new production line and the premium adjusts. Depreciation is fixed — until you buy the machine Nothing fancy..
And some "fixed" costs are actually step costs in disguise. Supervisory salaries. That's why you need one supervisor for every 15 workers. But that's a step function, not a flat line. But inside a narrow range — say, 10 to 14 workers — it looks fixed Surprisingly effective..
You'll probably want to bookmark this section.
Model it as fixed inside that range. Just know you're approximating.
Variable costs aren't always variable either
Direct materials? Direct labor? In practice, usually linear. Often linear — until overtime hits.
But utilities? Maintenance? Shipping? These love to curve.
Electricity has a base charge plus a per-kWh rate that drops at higher tiers. Because of that, maintenance is low and steady until equipment ages, then spikes. Shipping costs per unit drop when you hit full truckload quantities.
The more you dig, the fewer truly linear variable costs you find.
Why It Matters / Why People Care
You might wonder: if nothing is perfectly linear, why does every budget, every CVP analysis, every break-even calculation assume it is?
Because approximation beats paralysis Worth knowing..
Decision-making needs a starting point
Try building a flexible budget with curved cost functions for 200 cost centers. You'll still be at it next year.
Linear models give you a workable framework. They're "good enough" for most short-term decisions inside normal operating ranges. The error from assuming linearity is usually smaller than the error from bad volume forecasts anyway.
But — and this is critical — you have to know where the approximation breaks It's one of those things that adds up..
The danger zone: extrapolation
Here's where companies get burned Surprisingly effective..
They build a cost model at 10,000 units. It works beautifully at 9,500 and 10,500. So they use it to evaluate a proposal at 25,000 units.
The model says: "Great news! Also, fixed costs spread over 2. Even so, 5x the volume. Per-unit cost drops 40%. This project is a goldmine Surprisingly effective..
Reality: The plant hits capacity at 18,000 units. Plus, 5x rates, and expedited shipping. Still, the next 7,000 require a $2M expansion, temp labor at 1. Per-unit cost rises Still holds up..
The linear model didn't just miss — it pointed the wrong way Easy to understand, harder to ignore..
When linearity assumptions kill projects
I've seen this play out in three scenarios repeatedly:
1. Make vs. buy decisions — Internal cost looks linear at current volume. Outsource quote is fixed per unit. At higher volumes, internal variable costs drop (learning curve, better utilization). The linear model says "buy." Reality says "make."
2. Special orders — "We have excess capacity, variable cost is $12, they'll pay $15, easy money." But the order pushes a bottleneck resource past its limit. Now you're paying overtime or outsourcing a sub-process. Real variable cost: $16. You lost money on every unit.
3. Product line profitability — Allocate fixed costs linearly across products. Kill the "unprofitable" line. Remaining lines now absorb more fixed cost. They look unprofitable next. Death spiral.
None of this happens if you respect the relevant range.
How It Works (or How to Do It)
So how do you actually determine when cost behavior is linear — and when it's not?
Step 1: Define your activity base
Cost behavior is always relative to something. Machine hours. Labor hours. Customer visits. Units produced. Square footage Easy to understand, harder to ignore. Still holds up..
Pick the wrong base, and linear costs look curved. Pick the right one, and curved costs straighten out.
Example: A hotel's housekeeping costs. Per occupied room? Here's the thing — pretty linear. Think about it: per day? Curved — weekends are heavier. Now, per guest? Messy — single vs. double occupancy.
The activity base should drive the cost. But not correlate. Drive.
Step 2: Identify the relevant range for each cost
Don't guess. Talk to operations Surprisingly effective..
- At what volume does the current lease expire?
- When does the current supervisor max out?
- What's the practical capacity of the bottleneck machine?
- When do volume discounts kick in with key suppliers?
- At what point does overtime become mandatory?
Document these thresholds. They're your linearity boundaries Not complicated — just consistent..
Step 3: Classify costs — but verify
Standard classification:
- Variable: Direct materials, piece-rate labor, sales commissions
- Fixed: Rent, salaries, insurance, straight-line depreciation
- Mixed: Utilities, maintenance, telecom
But verify. Overages are $75/ticket. It caps at 500 tickets/month. Now, that "fixed" IT support contract? That's mixed — with a step Nothing fancy..
That "variable" sales commission? In practice, it has a tiered structure: 5% up to $1M, 7% above. That's piecewise linear — two different lines.
Step 4: Use the right estimation method
If you need to estimate cost functions from data (not just classify known costs), you have options:
High-low method — Uses only two data points. Fast. Terrible if those points are outliers. Don't use it for anything material.
Scattergraph method — Plot the data. Eyeball the line. Subjective, but you see non-linearity. Outliers jump out. Curves reveal themselves.
Regression analysis — The gold standard. Uses all data points. Gives you R-squared (goodness of fit), p-values (significance), and confidence intervals.
But regression assumes linearity. If the true
But regression assumes linearity. If the true relationship is curved, the model will still force a straight line, producing biased coefficient estimates and unreliable predictions. The first safeguard is to let the data speak before you trust the numbers.
Check the residuals. Plot the residuals (the difference between actual cost and the regression‑predicted cost) against the fitted values or the activity base. Randomly scattered residuals around zero indicate a good linear fit. Systematic patterns—wiggles, funnels, or clusters—signal that a straight line is not capturing the cost behavior. A residual plot is often more revealing than the regression statistics themselves.
Look for breakpoints. Many cost drivers exhibit “step” behavior. A supplier’s volume discount, a shift premium, or a capacity ceiling creates a change in slope at a specific activity level. Segmented (or piecewise) regression lets you estimate separate linear equations for each segment, preserving accuracy on either side of the breakpoint. The thresholds you identified in Step 2 become the natural cut‑points for these segments.
Transform the data when appropriate. If the cost curve is inherently curvilinear, a simple transformation can linearize it. Common tricks include:
- Log‑log regression for costs that scale proportionally with activity (power‑law relationships).
- Log‑linear regression when the cost grows exponentially with volume.
- Square‑root or inverse transformations for diminishing‑returns patterns.
Always test the transformed model’s residuals to confirm that the linearity assumption now holds Took long enough..
Use qualitative judgment alongside quantitative tools. Even the most sophisticated regression can be fooled by a single outlier—perhaps a one‑off maintenance repair or a temporary overtime spike. Operations managers can flag such events, allowing you to either exclude them from the analysis or treat them as separate cost elements Worth keeping that in mind. And it works..
Document the relevant range for each cost driver. Once you have a reliable cost function, annotate the activity levels at which it is valid. If a future forecast pushes volume beyond that range, revisit the assumptions: new lease terms, additional supervisors, capacity expansions, or revised supplier contracts may alter the cost structure.
Putting It All Together
- Define the activity base that truly drives the cost.
- Map the relevant range for each cost, noting capacity limits, contract thresholds, and overtime triggers.
- Classify costs but verify with operational knowledge—many “fixed” or “variable” items are actually mixed or step‑wise.
- Choose an estimation method that matches the data’s behavior: regression for a solid linear fit, scattergraph for quick visual checks, or segmented regression when breakpoints exist.
- Validate the model through residual analysis, transformation tests, and managerial input.
- Maintain the range boundaries in your cost models so that forecasts stay within the domain where the relationships hold.
By respecting the relevant range and applying a disciplined, evidence‑based approach to cost behavior, managers can avoid the death spiral of misallocated fixed costs, keep product lines truly profitable, and make decisions that reflect how costs actually move with activity. The goal is not to find a perfect straight line in a world of curves, but to understand where those curves straighten out and to model them accordingly Which is the point..
