The average rate of change of a function over an interval [a, b] is (f(b) − f(a)) / (b − a): the change in output divided by the change in input. That single ratio is the slope of the secant line joining the two endpoints. The condition attached to it matters just as much — the formula requires b ≠ a and tells you nothing about behavior inside the interval. A function can climb sharply, flatline, and fall inside [a, b] and still produce the same average as a straight line between the endpoints.
That gap between the interval's average and what actually happens inside is where most confusion lives. The articles that rank for this keyword usually give the formula, one worked example, and a promise that the average rate of change is "like slope." They skip the pieces students actually lose points on: the distinction from instantaneous rate, why the mean value theorem's guarantee depends on continuity and differentiability, how average speed can be nonzero while average velocity is zero, and what units the answer carries. This article covers the formula you came for, then the edge cases and computations that make it usable on an exam.
The Formula Stated Precisely, With Its One Hard Condition
The average rate of change of f over [a, b] is written as (f(b) − f(a)) / (b − a). Some textbooks present it as Δy/Δx, which is the same quantity with friendlier symbols: Δy is the change in the function's value, Δx is the change in the input. The order of subtraction must match on top and bottom. Computing f(a) − f(b) over b − a negates the whole answer and is the single most common arithmetic slip in this topic.
The hard condition is b ≠ a. If the endpoints coincide, the denominator is zero and the ratio is undefined — there is no interval to average over. A function's value at a single point, f(a), cannot produce a rate of anything on its own.
Two equivalent forms show up in coursework. The compact form is (f(b) − f(a)) / (b − a). The expanded form, used when a point is labeled with a variable, is (f(a + h) − f(a)) / h, where h is the interval's width. Setting h = b − a converts one into the other, and this second form is the one that becomes the derivative when h shrinks toward zero.
State the interval when you present an answer, not just the number. "The average rate of change is 3" is incomplete; "the average rate of change over [1, 4] is 3" is an answer. If the interval is open, (a, b), the average rate of change over it is still computed from the endpoint values, because open and closed intervals share endpoints. No rule prevents that computation, and textbooks treat the two identically for this purpose.
Worked Example: A Polynomial Over a Named Interval
Take f(x) = x² − 3x + 2 and compute the average rate of change over [1, 5]. First evaluate the endpoints. f(1) = 1 − 3 + 2 = 0. f(5) = 25 − 15 + 2 = 12. The numerator is 12 − 0 = 12, the denominator is 5 − 1 = 4, and the answer is 12/4 = 3.
Check the sign logic. The function rose from 0 to 12 across an input span of 4, so a positive average of 3 is consistent with an overall increase. Had the endpoints been reversed — from 5 to 1 — the same computation would give −3, because input now decreases. Direction is part of the answer.
Now take a quadratic with a turning point inside the interval. For g(x) = (x − 3)², the average rate of change over [1, 7] is (g(7) − g(1)) / (7 − 1) = (16 − 4) / 6 = 2. Yet g falls to 0 at x = 3 and climbs back. The average of 2 describes the chord from (1, 4) to (7, 16) and says nothing about the dip between them. This is the exact reason the average rate of change is not a description of the function's journey.
Try a third case where the function is not monotonic at either end. For h(x) = x³ − 6x over [0, 3]: h(0) = 0, h(3) = 27 − 18 = 9, and the average is 9/3 = 3. The function rises to a local maximum near x = √2 ≈ 1.41, falls to a local minimum near x = √2 ≈ 1.41 — sorry, the local minimum is near x = √2 on the other side, at approximately x = 1.41 for the maximum and x = −1.41 for the minimum within the relevant range — so within [0, 3], the local max is at x = √2 ≈ 1.41 and there is no local minimum on that half. Recompute carefully: h′(x) = 3x² − 6, zero at x = ±√2. On [0, 3] the critical point is x = √2 ≈ 1.414, a local maximum. The average of 3 still stands, and it hides a peak-and-decline segment near the start.
Every one of these examples carries a unit if the function does. If f(t) gives meters after t seconds, the average rate of change is in meters per second. Attaching units at each step prevents the common error of reporting a bare number to a physics question.
Secant Lines, Tangent Lines, and What the Average Cannot Tell You
The average rate of change is the slope of the secant line — the straight line through (a, f(a)) and (b, f(b)). On a graph, draw the chord and measure its rise over run. That geometric reading is exact, not an analogy: the secant slope equals the ratio by definition.
A tangent line touches the curve at one point and has slope equal to the instantaneous rate of change, the derivative. The two lines answer different questions. The secant answers "on average, how fast did the output change across this span?" The tangent answers "how fast is it changing right now?" A function can have a huge average over an interval and a zero instantaneous rate at every point inside it — a sawtooth wave does exactly this.
Here is the failure case worth remembering. A car drives 60 miles in 1 hour, stops for 30 minutes, then drives 60 more miles in 1 hour. Over the full 2.5 hours, average speed is 120/2.5 = 48 miles per hour. Over the middle 1.5 hours (from the end of the first leg to the end of the second), the car covers 60 miles in 1.5 hours, an average of 40 miles per hour. Neither figure ever equals the car's speedometer reading for more than a moment, and the speedometer reads zero during the stop. The average is not a "typical" speed; it is the constant speed that would cover the same distance in the same total time.
Slope sign carries a clean reading: positive average means net increase across the interval, negative means net decrease, zero means the endpoints have equal outputs. Zero is not the same as "no change occurred." A function that goes from 5 at x = 0 to 9 at x = 2 and back to 5 at x = 4 has an average rate of change of 0 over [0, 4] despite moving through a range of 4 units along the way.
Average Versus Instantaneous Rate: The Difference That Decides Exam Questions
The average rate of change is computed from two data points. The instantaneous rate of change is computed as a limit: the average as the gap between the two points shrinks toward zero. In symbols, the instantaneous rate at a is the limit of (f(a + h) − f(a)) / h as h → 0. That limit, when it exists, is the derivative f′(a).
For f(x) = x², the average rate over [1, 3] is (9 − 1) / 2 = 4. The instantaneous rate at x = 2 is f′(2) = 4 as well, and that coincidence is the mean value theorem working. Try the same function over [1, 4]: the average is (16 − 1)/3 = 5, and the instantaneous rate equals 5 at x = 2.5, the midpoint of the interval. The average is achieved somewhere inside, and the location is not always the midpoint — that's specific to quadratics.
The table below isolates the differences that exam questions probe.
| Feature | Average rate of change | Instantaneous rate of change |
|---|---|---|
| Inputs required | Two endpoint values, a and b | One point, a, plus the function's rule |
| Computation | (f(b) − f(a)) / (b − a) | lim h→0 (f(a + h) − f(a)) / h |
| Geometric meaning | Slope of the secant line | Slope of the tangent line |
| Units | Output units per input unit | Output units per input unit |
| Existence | Always defined for b ≠ a on a real-valued function | Exists only where the derivative exists |
| Effect of a corner | No problem; endpoints suffice | Undefined at the corner |
The table's last row matters more than it looks. At x = 0 on f(x) = |x|, the derivative does not exist — the graph has a sharp corner and no single tangent slope. The average rate of change over [−1, 1] is still perfectly computable: (|1| − |−1|) / 2 = 0. Averages survive corners; instantaneous rates don't.
The Mean Value Theorem and the Guarantee It Buys
The mean value theorem says that if f is continuous on [a, b] and differentiable on (a, b), then some c in (a, b) satisfies f′(c) = (f(b) − f(a)) / (b − a). In plain language, under those conditions the instantaneous rate equals the average rate at least once inside the interval. The secant line and a tangent line are parallel somewhere.
Both hypotheses are load-bearing. Drop continuity: consider f(x) = 1/x on [−1, 1]. The function is not continuous at 0, and there is no c where f′(c) equals the average (−1 − 1)/2 = −1, because f′(x) = −1/x² is always negative and takes no value in the theorem's sense at a finite point in the way required. Drop differentiability: f(x) = |x| on [−1, 1] is continuous but not differentiable at 0, and the average rate of change is 0 — no interior point has derivative 0 because the derivative never equals 0 and doesn't exist at the corner.
The practical value: on an exam, the mean value theorem is the justification tool. If a problem asks "must the instantaneous rate equal the average somewhere?", the answer is yes when continuity holds on the closed interval and differentiability holds on the open one. If the question says "at the midpoint," the answer is no — the theorem guarantees some c, not c = (a + b)/2. The midpoint coincidence appears for quadratics and vanishes for cubics and above.
Check the theorem's number for f(x) = x³ on [0, 2]. Average: (8 − 0)/2 = 4. Solve 3c² = 4 for c = √(4/3) ≈ 1.155, which lies inside (0, 2). Compare with the midpoint 1, which gives f′(1) = 3, not 4. The theorem earns its place precisely because it names a location the midpoint rule misses.
Reading the Sign and the Size: Interpretation Rules That Hold
Three rules convert the numeric answer into English. Positive answer: output increases across the interval, on net. Negative answer: output decreases on net. Zero answer: endpoints match. The fourth rule is the one exam graders watch for — a small magnitude does not mean the function is nearly constant. An average of 0.1 over a long interval can hide wild swings inside.
Size depends on units, so compare magnitudes only within the same unit system. A population model in people per year and a distance model in meters per second cannot be ranked against one another. Within one model, an average of 3 meters per second is faster than an average of 2 meters per second over comparable intervals, and that's all the comparison licenses.
Interpret the numerator and denominator separately before dividing. For V(t) = volume in liters after t minutes, an average of −2 liters per minute over [0, 5] means the volume dropped 10 liters across 5 minutes. Splitting the change from the span catches errors like dividing by the wrong endpoint difference. That split also gives a quick sanity check: multiply your answer back by the interval width and confirm you recover the numerator.
Conditional warnings belong here too. If the interval includes a discontinuity of the function itself, the average may exist yet have no physical meaning — an object cannot teleport. When the function models a real quantity, check that the endpoints and everything between are physically admissible before applying the number.
Real-World Applications: Velocity, Revenue, and Population
Average velocity equals displacement divided by elapsed time. If a runner's position is s(t) = 3t² + t meters after t seconds, the average velocity over [2, 4] is (s(4) − s(2))/(4 − 2) = (52 − 14)/2 = 19 meters per second. Note that this is displacement-based; average speed, which averages the distance traveled, can differ when the motion reverses.
Revenue questions work the same way. If a company's monthly revenue was 40,000 dollars in month 3 and 64,000 in month 9, the average rate of change is 24,000/6 = 4,000 dollars per month. That is the uniform monthly growth that would connect the endpoints, and it is not a forecast. A nonlinear revenue path could hit 4,000 dollars in a month at one point and dip in another.
Population models behave the same. A town of 12,000 in 2015 and 15,600 in 2023 has an average growth of 3,600/8 = 450 people per year. If the town hall reports a different "typical" annual increase, the average is still 450 for those endpoints by arithmetic — the disagreement is about which metric was reported, not the computation.
The cross-discipline trap is unit laundering. In velocity problems, the interval width is time and the numerator is distance; in revenue, the width is months and the numerator is dollars. Reporting "4,000 per month" as "4,000 per year" multiplies the figure by twelve in the reader's mind. Always write units in full and check that they survived the division.
Common Mistakes, Edge Cases, and How to Catch Them
Endpoint order: computing f(b) − f(a) over a − b flips the sign. Do both differences in the same direction and never mix. Students who get "the wrong sign" almost always flipped one of the two subtractions.
Forgetting the width: using only f(b) − f(a) drops the denominator and turns a rate into a change. The change is in output units; the rate is in output units per input unit. If your answer's units are wrong, this is why.
Assuming the function is linear inside: the average of an interval is not a description of interior behavior. The quadratic dip example earlier is the antidote — average 2, dip to zero, climb back. State the average and say it is the chord slope.
Confusing open and closed intervals: students sometimes refuse to compute a rate over an open interval. Both interval types give the same average because both use the endpoint values, and no rule forbids it.
Failing to check continuity for the mean value theorem: if the function has a jump inside [a, b], the theorem's conclusion can fail even though the average computes normally. The failure case with 1/x on [−1, 1] is the standard counterexample, and it is worth memorizing.
For functions given as a table, interpolate only if the problem says the table fully represents the function. Otherwise the average rate of change over [a, b] uses the two listed endpoint values, and midpoints are irrelevant.
FAQ
What is the average rate of change of a function over an interval in one sentence?
It's the slope of the secant line through the interval's endpoints, computed as (f(b) − f(a)) / (b − a). The result carries the same units as output per input. If the endpoints share an output value, the answer is zero even if the function moved in between.
Can the average rate of change be negative, and what does that mean?
Yes, and a negative answer means the output is lower at the right endpoint than at the left. For a position function, a negative average over an interval means net backward displacement across that span. The size reflects how fast the net decrease was, not how any single moment looked.
Is average rate of change the same as slope?
It is the slope of the secant line, which is a slope but not the curve's slope at any point. Textbooks often say "average rate of change is slope" as shorthand for that geometry. The curve's own slope at a point is the instantaneous rate, a different quantity. Triangle problems about velocity force the distinction.
What happens if the two interval endpoints are the same?
The formula fails because the denominator b − a is zero. There is no interval and no rate. Define b ≠ a before computing, and if a problem gives you one point, that's a cue that the question is asking for the instantaneous rate instead.
Does the average rate of change always equal the instantaneous rate somewhere inside?
Only when the function is continuous on [a, b] and differentiable on (a, b), which is exactly what the mean value theorem requires. Even then, the theorem promises some point, not the midpoint. For f(x) = x³ over [0, 2], the location is about 1.155, not 1.
How do I compute the average rate of change from a table of values?
Pick the two rows whose inputs are the interval's endpoints and apply (output2 − output1) / (input2 − input1). Do not average the individual rates between successive rows — that weighted average is only correct for equally spaced inputs and simple interpolation. The endpoints are all you need for the interval average.
What's the difference between average speed and the average rate of change of position?
Average speed divides total distance traveled by time; the average rate of change of position divides net displacement by time. For a round trip returning to the start, displacement is zero so the position average is 0, while average speed remains positive. Use the position average when direction matters.
Why does my calculus class compute this with limits?
The limit process converts an average over a shrinking interval into the instantaneous rate, the derivative. The formula you compute with a limit is the same formula with b replaced by a + h and h approaching zero. Once you have the derivative, you can evaluate the instantaneous version anywhere it exists.