The average rate of change from a table is the change in output values divided by the change in input values between two rows: (y₂ − y₁) ÷ (x₂ − x₁). The average spans the whole interval between the endpoints — it is not the slope at any single point — so the only values you need are the two y-values and the two matching x-values. Read those four numbers off the table first, subtract in the same order on top and bottom, and handle a vertical-line table (equal x-values) as undefined rather than as zero.
What the Average Rate of Change Means in a Table of Values
A function table lists paired inputs and outputs. The average rate of change between any two of those rows measures how fast the output moves per one unit of input, averaged across the whole gap between the rows. If a table shows a car's distance at 2 p.m. and again at 4 p.m., the average rate tells you the steady speed that would have covered the same ground in the same time. It cannot tell you the speed at 3 p.m. — that instantaneous rate is a different calculation and needs a graph or a formula.
The word "average" does real work here. A table row records only the moment at the measurement, not what happened in between. A cyclist who sprints for ten minutes and coasts for fifty can share the same average rate as one who pedals steadily for an hour, provided the start and end readings line up. That's a feature, not a flaw: average rate of change answers "how much, per unit of input, over this stretch," and it answers it whether or not the underlying pattern is smooth.
Tables come in two layouts, and the layout changes which column you subtract first. In a horizontal table, the input row sits on top and outputs below it; in a vertical table, input is usually the left column. Either way, the change in the dependent quantity goes in the numerator. Mixing up which quantity belongs on top is the most common arithmetic slip in this topic, and it flips the sign of the answer whenever the function is decreasing.
One boundary worth stating up front: the result depends on the two rows you pick. Change the endpoint pair and the average rate changes. That isn't a bug in your work — it's how the quantity is defined, and it's why calculators ask "between which points" before they ask anything else. When a problem says "find the average rate of change," it means between the rows you're given, or between the first and last rows if none are specified.
The Formula and the Division Order That Keeps Signs Correct
The formula is (f(b) − f(a)) ÷ (b − a), where a and b are the two input values and f(a), f(b) the matching outputs. Subtract in the same direction on top and bottom: if b is the later input, it comes first in both places. Reversing one subtraction but not the other gives the right magnitude with the wrong sign, which is how a falling price series ends up reporting a positive rate.
Because division order is everything, write the two coordinates as ordered pairs before subtracting. If a table gives (3, 20) and (7, 44), then f(b) − f(a) is 44 − 20 = 24 and b − a is 7 − 3 = 4, so the average rate of change is 24 ÷ 4 = 6. Run the same pair backwards — (7, 44) first, (3, 20) second — and you get (−24) ÷ (−4) = 6. Same answer, because both subtractions flipped. Flip only one and you get minus 6.
A single convention covers nearly every case: name the left-hand (earlier) point a and the right-hand (later) point b, then compute. Some textbooks call the numerator Δy and the denominator Δx, read aloud as "delta y over delta x." Delta just means "change in," so Δy is the difference between the output values and Δx the difference between the input values — nothing more mysterious than that.
The formula also works when the inputs run downward on the page. Some tables list years in descending order, or measurements from latest to earliest. If a = (2020, 150) and b = (2018, 120), then Δy = 120 − 150 = −30 and Δx = 2018 − 2020 = −2, giving a positive rate of 15 per year — correct, because the quantity grew from 2018 to 2020. Pick your endpoints by value, not by row position.
Step-by-Step: Reading Two Rows and Computing the Result
This is how I'd work through any table problem, in the order you'd actually write it on paper. The sequence matters: you can't subtract reliably until you've labelled which row is which.
1: Name the two points as (x, y) pairs. Write them side by side, first point on the left. Don't skip this — writing the pairs prevents the sign error in step 3 from ever happening. 2: Decide which point is a and which is b. Use the smaller x for a, the larger x for b, unless the problem names specific rows. 3: Compute Δy = y_b − y_a. Keep the order: b's output minus a's output. 4: Compute Δx = x_b − x_a. Same order, b's input minus a's input. 5: Divide Δy by Δx. Reduce the fraction if it lands on something like 12/8, but keep the decimal too — 1.5 is easier to compare against other intervals. 6: Attach units. Output units per input unit: dollars per year, litres per minute, kilometres per hour. A bare number like 6 usually earns partial credit because the units went missing.
If the problem gives more than two rows and asks for the rate "over the whole table," use the first and last rows only. Walking every adjacent pair and averaging those results gives a different number, and it isn't what's being asked — each adjacent pair is its own interval with its own rate.
Here's the failure handling. If Δx comes out as zero, stop: the two rows share an input value, the line is vertical, and the average rate of change is undefined — not zero, not infinity, undefined. Write "undefined" and say why. If Δy is zero while Δx isn't, the rate is exactly 0, which means the output didn't move across that interval; that's a perfectly valid answer and a different animal from undefined. And when the table carries units in the header (thousands of dollars, thousands of people, years since 2010), convert before subtracting or state clearly that the rate is in thousands, because a rate of 4 on a table measured in thousands is really 4,000.
Worked Example One: A Linear Table Where Every Pair Agrees
Start with the friendliest case — a table generated by a straight line — because it builds the checking habit. Take inputs 0, 2, 4, 6 with outputs 5, 13, 21, 29. Between rows one and four: Δy = 29 − 5 = 24, Δx = 6 − 0 = 6, so the average rate of change is 4. Between the first two rows only: 13 − 5 = 8 over 2 − 0 = 2, which is also 4. Every pair of rows reports the same rate, and that's the fingerprint of a linear function — constant slope.
That agreement is your built-in check. When a table is linear, any endpoint pair must return the same number, so if one pair gives 4 and another gives 5, you either misread a cell, subtracted in mixed order, or the function isn't actually linear. Working the check costs about fifteen seconds and catches nearly every arithmetic slip before it reaches the page.
For a steeper comparison: inputs 1, 3, 5 with outputs 10, 26, 42. Between the outer rows, Δy = 42 − 10 = 32 and Δx = 5 − 1 = 4, a rate of 8. Between the first two, 26 − 10 = 16 over 3 − 1 = 2, again 8. Notice that as the input grows by a larger step, the output jumps by the same proportion — 4 input units buy 32 output units, 2 buy 16 — because a straight line scales everything evenly.
The units are where students lose marks. If the input column is labelled "months" and the output "subscribers," the answer is 8 subscribers per month, not 8. If both columns are in thousands, the honest statement is "8 thousand subscribers per month." Deriving the units matters more than any single number here: per-unit rates always carry output units on top and input units underneath.
One last linear-table quirk worth knowing: reading a slope directly off a table with a step larger than 1 means dividing by that step. A table that jumps x by 4 each row shows a y jump of 32, and 32 ÷ 4 is 8 — not 32. Skipping the division is the single most common error when the x-column steps are more than one.
Worked Example Two: A Nonlinear Table Where the Average Depends on the Interval
Real tables rarely cooperate. Take a population table: year 2010 with 12,400 residents, 2015 with 14,900, 2020 with 16,800. Between 2010 and 2020, Δy = 16,800 − 12,400 = 4,400 and Δx = 2020 − 2010 = 10, so the average rate is 440 residents per year across the decade. Between 2010 and 2015 alone: 2,500 ÷ 5 = 500 per year. Between 2015 and 2020: 1,900 ÷ 5 = 380 per year.
The three numbers disagree — 440, 500, 380 — and all three are correct. Each describes its own stretch. The town grew faster in the first half of the decade than the second, so the decade-wide average sits between the two half-decade rates. This is exactly what "average" means for a nonlinear function, and it's why the interval must always be stated alongside the answer.
Now the valuable part: you can compute the decade rate without ever looking at the 2015 row. The middle row is irrelevant to the 2010–2020 answer. Students often try to fold every row into the calculation, summing all the changes and dividing by the count of gaps, which produces a number that answers no question anyone asked. The average rate of change uses two rows — always two, never all.
A decreasing version keeps the sign honest. Suppose a car's value table reads age 1 with $24,000 and age 4 with $15,000. Δy = 15,000 − 24,000 = −9,000; Δx = 4 − 1 = 3; rate = −3,000 dollars per year. The negative sign tells you the value is falling at an average of $3,000 annually across those three years — an average, mind you, not a guarantee that it fell exactly $3,000 in any single year.
Tables Where the Average Rate Is Undefined or Misleading
Two rows stacked at the same input value produce a real problem. If a table shows x = 5 with y = 20 and another x = 5 with y = 27 — possible when a survey records two readings at the same time, or when a table lists a vertical jump — then Δx = 5 − 5 = 0 and division has no defined result. The average rate of change is undefined. Writing zero is a conceptual error, not a rounding issue: zero would claim the output didn't change, and it plainly did.
The opposite zero deserves equal care. If both rows show the same output, Δy = 0 and the rate is 0 per unit of input, provided Δx isn't also zero. That's a legitimate flat stretch — a month with no price change, a plateau between measurements.
A second trap is the equal-spaced assumption. A table listed at 2, 5, 4 and 8 in the x-column looks tidy but isn't evenly spaced; each adjacent gap differs. Compute only the pair you need, and if a question asks for the rate across the whole table, the first and last rows must come from the actual endpoints, not from whichever rows are physically uppermost.
Rounding counts as a third failure mode. A rate of 2,500 ÷ 3 = 833.33… shouldn't be reported as 833 if the next question compares intervals, because the small truncation compounds. Keep the fraction or carry two extra decimals, and round only in the final sentence.
Rate of Change vs Slope vs Instantaneous Rate
Three terms get braided together in textbooks, and untangling them stops most confusion. Average rate of change is the change in output divided by the change in input across an interval — the table calculation covered here. Slope is the same number when the function is linear, because a straight line has one slope everywhere and the average rate across any interval matches it. For a curve, the average rate is a chord, not a slope at a point.
The instantaneous rate of change is the limit of the average rate as the interval shrinks toward zero. Tables can only approximate it: the average rate between 4 and 4.1 seconds is closer to the instantaneous rate at 4 than the average between 4 and 8 seconds, and tables of values are usually built to let students shrink the gap. On a distance–time table, average rate of change is average speed and the instantaneous rate is the speedometer reading at an instant.
The units test separates them cleanly. Both average and instantaneous rates carry output-per-input units. Slope is unitless as a geometric concept but gains units when the axes carry them — a line plotted with "cost in dollars" on the vertical and "weeks" on the horizontal has a slope of dollars per week, identical to the average rate of change. If someone asks for the slope and you have a table, compute the average rate; for a linear table the two numbers match.
A quick numerical demonstration of the shrinking interval: for a falling object, the distance table might read 1 second with 4.9 metres and 1 second later with 19.6 metres. The average rate over that second is 14.7 metres per second. Halve the interval to a tenth of a second and the average rate inches toward 9.8 — the instantaneous rate at 1 second, which is the acceleration due to gravity, 9.8 metres per second squared. Students who spot that 9.8 will remember the difference between the two rates for good.
Checking Your Answer Against the Table and Real Units
Good practice is a reverse-engineered check: take your rate, multiply by the input gap, and confirm you land on the output gap. If the reported rate is 6 and the input gap is 4, then 6 × 4 = 24 must equal the y-difference. If it doesn't, one of the two subtractions is wrong. This check takes one line and rules out most mistakes.
Unit sanity is the second filter. A factory table counting widgets per hour should not produce a rate in the thousands unless the factory is enormous; a rate of 0.02 widgets per hour at a small operation is plausible. If the number looks absurd, check whether a header note says "in thousands" or "in millions," because tables routinely scale their entries and hide the note in the caption.
The third filter is direction. Rising output pair with a positive input gap gives a positive rate; falling output gives negative. If your answer contradicts the arrows in the table, reverse one subtraction. A quick physical read — the values clearly go down from row to row — settles the sign in half a second.
Finally, when a question gives no interval, state yours. An answer of "6" without "between the 1995 and 2005 rows" is incomplete, and the same table can legitimately yield a different rate for every pair of rows.
Choosing Which Rows to Subtract When the Question Is Vague
Many practical questions arrive without a specified interval: "what's the growth rate of this table?" The professional answer names the endpoints you chose before giving the number. Most curricula default to the first and last rows when a table is presented whole, because the endpoints bracket the entire record and the rate then summarizes the full period, like a decade of census growth.
But context can override the default. A table of sales that jumps when a new store opened might better be split at that row, with separate rates before and after; the question "how fast are sales growing" then has two honest answers. If a table has missing rows or a gap, swapping the input pair changes the answer, so use only measured rows and never interpolate a value to manufacture an endpoint.
Tests usually resolve the ambiguity, or they expect the full-table endpoints. Outside tests, the safe habit is: state endpoints, state unit rate, and, if the function looks curved, add a sentence noting the rate would differ for a narrower interval. That one sentence kills almost every follow-up objection.
A quick closing example of the whole routine. The table lists a plant's height in centimetres at days 2, 5, 9: 6, 12, 21. Between days 2 and 9, Δy = 21 − 6 = 15, Δx = 7, average rate = 15/7 ≈ 2.14 centimetres per day. Between days 2 and 5 it's 6 ÷ 3 = 2 cm/day; between 5 and 9 it's 9 ÷ 4 = 2.25 cm/day. The full-interval rate sits between the two adjacent rates, the way a fair average should — and all three answers are correct statements about three different windows.
FAQ
What is the average rate of change if both y-values are the same?
The rate is exactly zero, because Δy = 0 and any nonzero Δx divides cleanly into it. That means the output held steady across the interval, not that the calculation failed. Keep the sign in mind: zero is neither positive nor negative growth. Only when Δx is also zero does the answer become undefined rather than zero.
Can the average rate of change be negative?
Yes, and the negative sign is information, not an error. A table where the later row's y-value is smaller than the earlier row's gives a negative Δy and, with a positive input gap, a negative rate. Flipping the subtraction order without flipping both gives a sign error — a car worth less each year reads −3,000 dollars per year, so the minus sign belongs in the answer.
Why is the average rate of change undefined when the x-values match?
Division by zero is undefined, and two rows sharing an input value produce exactly that: Δx = 0. Geometrically, the two points stack vertically, and a vertical line has no defined slope. Writing "undefined" and naming the shared x-value is the full credit answer; writing "0" or "infinity" both earn none.
Does the answer depend on which two rows I pick?
Always for a nonlinear table, and the endpoint pair must be stated alongside the answer. A population that grows 500 per year in one half-decade and 380 in the next reports 440 across the full decade — every one of those numbers is right for its own window. For a linear table every pair agrees, so the question of which rows barely matters.
How is this different from finding the slope?
For a linear table they're the same number, and slope is a perfectly good name for the answer. For a curve, the average rate of change is the slope of the chord connecting two points, not the slope at a single point — that's the instantaneous rate of change, which needs a limit and can only be approximated from a table by shrinking the gap.
What units should the final answer carry?
Output units per input unit, always. A table with metres on the output and seconds on the input yields metres per second; dollars per year, litres per minute, and widgets per month follow the same pattern. If the column header says the values are in thousands, the rate inherits that scaling and must say so — a bare 8 on a table measured in thousands means 8,000.
How do I quickly catch arithmetic mistakes?
Multiply your rate by the input gap and check the result equals the output gap. A rate of 6 across an input gap of 4 must reproduce a y-difference of 24; if it doesn't, reverse one of your subtractions. For linear tables, recompute with a second pair of rows — any disagreement between the two results flags a misread cell or mixed subtraction order.
Can I compute the rate from rows that aren't the first and last?
Yes, any two rows in the table support their own average rate, and upper-level questions often ask for a specific pair or for the rate between two named rows. What no question accepts is averaging the row-to-row rates, or summing them, to stand in for an interval rate — that answers a different question.