A table represents a linear function only when the change in output divided by the change in input stays identical between every consecutive pair of rows. If that ratio shifts anywhere, the table is not linear, no matter how tidy the numbers look. The test takes about a minute per table, and this article walks through it with worked examples, the vertical-line reading of x = 3, the discrete-versus-continuous distinction, and the two arithmetic traps that produce false positives on short tables.

Subtract Consecutive Outputs And Inputs To Build A Difference Table

Start by adding two extra columns to the right of whatever data you were handed: one for the change in x and one for the change in y. For any two rows, delta-x is x₂ minus x₁ and delta-y is y₂ minus y₁; the order matters because a reversed subtraction flips every sign and makes a rising function look like a falling one. A typical worksheet row like (2, 7) followed by (4, 11) gives a change in x of 2 and a change in y of 4. Fill those columns all the way down before you compute a single slope. Doing the arithmetic in two passes stops you from alternating between subtracting inputs and dividing them in your head.

For the table below, the completed difference columns look like this:

x y Δx Δy
1 5 — —
3 11 2 6
5 17 2 6
7 23 2 6

Notice that the y-values 5, 11, 17, 23 climb by 6 each time while the x-values climb by 2. Those two running totals are the raw material for the slope test, and on a clean linear table they stay constant from row to row. If either column of differences varies, your slope will vary with it. Keep the table's original order; re-sorting a scrambled table can create the illusion of a pattern that isn't there.

Pay attention to sign as you fill the columns. A y-value dropping from 20 to 14 gives a delta-y of negative 6, and that minus sign has to survive into the division step. Students who drop it end up with a negative slope on a function that's actually rising, then reject a correct linear table. Write the sign down even when it's negative, and keep the x-column signs consistent the same way.

One more practical tip: if the x-values jump unevenly (1, 2, 5, 9), the delta-x column will not be constant, and that's fine. Linearity cares about the ratio, not the step size. You can read the difference table in either direction and get the same slope magnitudes, which is a useful cross-check when a long table has more than a dozen rows and you've lost track of the arithmetic.

Divide Each Output Change By Its Input Change And Compare Every Slope

The slope between any two consecutive rows is delta-y divided by delta-x, and the definition of a linear function is that this quotient never changes. In the table above, every pair of rows gives 6 ÷ 2 = 3. So the function rule is y = 3x + 2: check row one by substituting x = 1, which gives 5, and check the last row by substituting x = 7, which gives 23. Both match, so the table is linear with slope 3. When you get the same quotient from every consecutive pair, the table represents a line and you can write its equation from any single row plus that slope.

Run the same procedure on (0, 4), (1, 7), (2, 10), (3, 13) and you'll get 3 ÷ 1 three times in a row, so the slope is 3 and the intercept is 4, giving y = 3x + 4. Now try (0, 4), (1, 7), (2, 12), (3, 13). The first pair gives 3, the second gives 5, and the third gives 1. Three different slopes mean three different lines, so the table is not linear. The fail is immediate and unambiguous: you don't need to check the fourth pair once a single pair disagrees. Make the disagreement your stopping condition on long tables.

A horizontal row pair such as (4, 9) followed by (4, 12) is a special case, because delta-x is 0 and division by zero is undefined. That table cannot be a function at all, because one input has two outputs, so the linearity question is moot. Vertical pairs are the reason the vertical line test exists, and you should report "not a function" before you report "not linear." It's a different error and teachers grade it differently.

The same extraction works backward from an equation. If a problem hands you y = −2x + 9 and a table of sampled points, just check that every row satisfies the equation. Two matching rows suffice in practice, but a third row catches copying errors in the table. Exact arithmetic from integers is what allows the no-estimating rule; all division here is exact, so you never need a calculator's decimal display and never need to round.

A Proportion Requires Equal Ratios Between Rows, Not Equal Differences Between Outputs

The table below separates a real linear function from the most common impostor on short datasets:

Table Pairs (x, y) Consecutive Δy Consecutive slopes Verdict
A (1, 2), (2, 4), (3, 6) 2, 2 2, 2 Linear, slope 2
B (1, 2), (2, 4), (4, 8) 2, 4 2, 2 Linear, slope 2
C (0, 1), (1, 2), (2, 5) 1, 3 1, 3 Not linear
D (1, 3), (2, 5), (3, 9) 2, 4 2, 4 Not linear
E (1, 3), (2, 5), (3, 7) 2, 2 2, 2 Linear, slope 2

Table A and Table B both represent y = 2x even though the x-steps differ, which proves that constant input spacing is not part of the definition. Table C and Table D both pass a casual "is it growing steadily?" eyeball check and fail the actual test, because their output jumps accelerate: C goes up by 1 then by 3, D goes up by 2 then by 4. Accelerating changes belong to quadratic or exponential behavior, not linear behavior. That's the single most common false positive on short tables.

The word "proportion" gets misused here, so pin it down. Proportional means the ratio y/x is constant and the line passes through the origin; the function y = 3x + 2 has a constant slope but is not proportional because 2 ÷ 1 is not 14 ÷ 5. If a word problem asks whether two quantities are proportional, dividing each y by its x is the right test and it is stricter than the slope test. Mixing the two tests is how students mark a proper additive relationship as proportional.

Table E shows the honest case: constant slope 2 with an intercept that isn't zero. It's linear and it's not proportional, and both statements are correct at once. Keep those labels separate in your head, and the rest of the classification work stays clean.

Spot the Linear Rules In Disguise When The Table Looks Messy

A table with irregular x-values can still be perfectly linear, and students who only ever see step-1 tables misjudge them. Test (2, 8), (5, 17), (9, 29): slopes are 3, 3, 3, so the underlying rule is y = 3x + 2 and the table is linear. The x-values jumped 3 then 4, while the y-values jumped 9 then 12, a ratio that holds at 3 both times. Print that pair of figures side by side before you complain about an uneven table, because the ratio is the only thing that matters.

The opposite disguise is the table with constant output. A function like y = 4 gives the same value in every row, so every delta-y is 0 and every slope is 0. That's a horizontal line, which is linear. Don't confuse it with an undefined slope from a repeated x-value, which is vertical and not a function. One zero sits in the numerator; the other sits in the denominator.

A line with slope zero and a line with slope one both deserve a quick sanity check. If your table's output never changes while the input does, the answer is a constant function with slope 0; if your slope comes out as 1, the input and output columns should be rising or falling in lockstep, and any row that breaks that lockstep points to a typo. These two speck-level cases show up on standardized tests because they catch students who only recognize slope values like 2 or −3.

Actual data tables rarely sit on exact integers. Denote your slopes as your "best estimate" when a table comes from a measurement with rounded values such as tenths or hundredths; small variation in the quotients is rounding noise, but a clean match to three decimal places is strong evidence of linearity. Report the evidence, not a verdict.

When The Domain Is A Sorted List Of Points, Linearity Is About The Underlying Function

A table of discrete data gives you inputs and outputs, but the phrase "represents a linear function" means there is a straight-line rule behind all of them, not just a straight-line pattern on paper. Two points, or a table with only two rows, is not enough to prove linearity: a line always fits two points, so you need at least three rows to observe a constant slope and a fourth row to trust it when the numbers were measured rather than generated. On a clean four-row integer table, a third slope agreement makes the case.

Beware a small table that is missing its middle row. Given (1, 2), (3, 8), (5, 14), the slope from row one to row two is 3 and from row two to row three is 3, so the function rule y = 3x − 1 fits every row and any unlisted input. Now delete the (3, 8) row and you're left with (1, 2) and (5, 14): the slope is still 3, but the table has no longer verified anything, because a single interval always agrees with its own slope. When a short table is missing an interior row, assume the linearity it shows in the gaps; state the assumption rather than dressing it up as proof.

Vertical Lines And Other Non-Functions Expose The Input-Output Rule

A vertical line test says: run your eye down the x-column, and if any x value repeats with a different y value, you're not looking at a function, so linearity testing is pointless. The classic table (3, 7), (1, 2), (3, 11), (4, 15) fails at the second and third rows, since x = 3 maps to both 7 and 11. That is the vertical line test failing at the simplest scale, and no slope arithmetic can rescue it. If you run the division anyway, two rows share an x and at least one delta-x equals 0.

The table format also has a reading trap. A table sorted by y rather than by x can look like a function while the x-values are scattered; sorting by y is convenient but it destroys the reading order. Restore the x-order before you test, and the vertical-line failure will usually pop out in the first two minutes. For a function rule that came out of a real table, "function" and "linear function" describe different properties, and a table can be one and not the other.

Tap The Finite-Differences Check With Wolfram|Alpha On Long Or Decimal Tables

For long tables, hand-checking every consecutive slope is slow and error-prone. Enter the pairs into WolframAlpha, or paste them into Desmos as an (x, y) list to view the scatter; Excel can compute each delta-y over delta-x in one formula column. In WolframAlpha, format the input as a set of ordered pairs and ask for the linear fit or the differences, and the tool reports what the data actually supports. It also accepts plain-English requests, which helps when you're cross-checking a child's answer. A line of best fit that has a nonzero residual means the points are not perfectly collinear, so a decimal table needs its fit tolerance stated.

Software doesn't settle the school question, though. Calculators comply with whatever arithmetic you feed them, and limited-precision tables produce slopes like 1.94, 1.98, 2.01 from rounding; a linear fit then splits the difference rather than certifying exact linearity. Treat a tool's R² near 1 as evidence, not proof, and reserve the word "linear" for tables where every consecutive quotient is identical at the precision given. When a grading step asks "does this table represent a linear function," the expected evidence is the paired slopes themselves.

Work A Short Training Set With One Reversed Row And A Decimal Slope

Here's a warm-up set to practice with. Table one: (6, 13), (8, 19), (10, 25), (12, 31). Differences are 6 each time and slopes are 3 each time, so the rule is y = 3x − 5 and the table is linear. Table two: (3, 1), (4, 3), (5, 7), (6, 13). Slopes are 2, 4, then 6, so this table accelerates and is not linear. Table three: (1, 0), (2, 3), (3, 6), (4, 9), the cleanest possible slope-3 line, with y-intercept 0, showing linear and proportional together. Table four: (1, 2.5), (2, 4.1), (3, 5.7), (4, 7.3). Differences are 1.6, 1.6, and 1.6, so the slope is 1.6 and the rule is y = 1.6x + 0.9; the decimals don't disqualify it.

Then reverse one row's inputs on purpose: (2, 5), (4, 9), (6, 15), (8, 19). The slopes are 2, 3, 2. That single displaced row is the reason you should sort by input before you start, because a mid-list slip of one y-value can masquerade as a "mostly linear" pattern that a single rule can't explain. Find the row where the slope breaks, remove it, and test the rest. If the remaining rows are consistent, you've found a typo rather than a genuine non-linear relationship, and the distinction matters when the table came from data entry rather than a textbook.

FAQ

What if the x-values aren't consecutive integers — does the table still count?

Yes, as long as the ratio, not the gap, stays constant; the function y = 3x + 2 can be sampled at x = 2, 5, and 9 and still pass with slopes of 3, 3, and 3. Uneven input spacing only means the Δx column varies, which is normal for real data. If you get a consistent ratio across uneven gaps, that's stronger evidence of linearity than an integer table, because the pattern can't be a coincidence of matching gaps.

How many rows does a table need before I can say it's linear?

At least three rows with matching consecutive slopes, ideally four; with only two points there is exactly one line through them, so a two-row table can't distinguish linear from curve. A fourth row guards against the "two slopes by luck" case on short tables. For measured data, add a fifth row if the values are rounded to whole numbers and you need reasonable confidence.

Is a table with a constant output column always a constant function with slope zero?

Only if the x-values differ; a constant y with the same x repeating isn't a function at all. With distinct inputs, a constant output is y = k, whose graph is a horizontal line, so it's linear with slope 0. That slope of 0 is a real value; it isn't a missing slope or an error from a blank division.

Which is the faster hand method, calculating slopes or calculating differences?

Differences are faster if the x-values are equally spaced, because equal input steps make constant output differences the same as constant slopes. Both readings work; the difference column and slope column give identical verdicts on equally spaced tables, at approximately the same effort. Save slope division for tables with uneven gaps, where differences can look constant while the ratios drift.

Can a table represent an exponential function and still be confused with linear, and how do I avoid it?

Yes: (0, 3), (1, 6), (2, 12) has constant x-steps and growing y-outputs, so the slopes 3 and 6 disagree and the table is exponential rather than linear. The distinguishing property is that linear differences are constant while exponential ratios y₂/y₁ are constant at 2 — check one property, not both, to avoid double-checking as if they were the same test. Linear tables grow by constant additions; exponential stacked patterns multiply.

Does every linear function have a constant positive slope?

No; slope can be negative, zero, or a fraction, and any fixed value is fine. The table (1, 10), (2, 7), (3, 4) has slopes of −3 and −3, which is linear with a negative slope and still passes the slope test. The word "linear" describes the straight-line shape, not the direction of the line.

What if the table is from a real experiment and slopes are close but not equal?

Close means rounding, and a sorted table with differences like 1.98 and 2.02 from tenths-precision readings usually sits within rounding error. Check the original source's precision before calling it non-linear; a measurement rounded to one decimal can shift a slope by 0.05 without any change in the relationship. Only report "not linear" when the gaps in consecutive ratios are well outside that noise band.