A table describes a linear function when subtracting the same input step produces identical output differences; it describes an exponential function when those subtraction answers keep changing but the ratios between consecutive outputs stay fixed. One test settles it: compute first differences and consecutive output ratios. Linear data has constant differences and varying ratios. Exponential data has constant ratios and varying differences. Three evenly spaced rows are enough to make the call β€” and the fourth row is where most students get fooled.

That paragraph is the whole method. What follows is the version you can actually run on a worksheet: how to build the difference and ratio columns by hand, what a $+50$ versus $\times 1.25$ pattern looks like on real tables, the domain quirks that break the test (a zero output kills every ratio, a population table hides a 0.5 person), and the fastest way to write the equation once you've made the call.

The difference test and the ratio test, run side by side on one table

Set up two working columns beside any table of values: one for first differences and one for consecutive ratios. A first difference is output minus previous output. A consecutive ratio is output divided by previous output. Both columns assume your input values are evenly spaced β€” steps of 1, steps of 5, steps of 0.5 all work, but a table that jumps from $x = 1$ to $x = 3$ to $x = 10$ breaks the difference test and leaves the ratio test meaningful only if you account for the uneven steps.

Run both columns on every table, even when the answer looks obvious. The columns are cheap to compute and they expose the noise: a linear table with one arithmetic slip shows up as three identical differences and one outlier, which tells you to recheck the row rather than declare the data exponential.

The decision rule needs three consecutive outputs minimum. Two points always fit a line and always fit an exponential curve, so a two-row table is unanswerable β€” a point worth remembering when a quiz hands you exactly two rows and expects a verdict. With three rows you get two differences and two ratios. Linear data produces two equal differences; exponential data produces two equal ratios.

Input Output First difference Consecutive ratio
0 6 β€” β€”
1 18 12 3
2 54 36 3
3 162 108 3

Here the differences run 12, 36, 108 β€” not constant, so nobody would call it linear. The ratios hold at 3 across every step, which makes the function exponential with base 3 and starting value 6: y = 6 Γ— 3^x. The difference column is not useless, though; in exponential data each difference equals the previous difference times the base, because 12 Γ— 3 = 36 and 36 Γ— 3 = 108. That relationship gives you a second confirmation of the base, and it's the detail most table walkthroughs skip.

Why constant ratios prove exponential and constant differences prove linear

The algebra behind the tests is short. A linear function adds the same amount every time the input rises by one, so f(x+1) βˆ’ f(x) equals a fixed slope m no matter which row you're in. An exponential function multiplies by the same factor every time the input rises by one, so f(x+1) Γ· f(x) equals a fixed base b. Each test is the defining property of its function family, measured directly off the table.

That matters because students often try to identify the function from its starting value or its shape, and both cues lie. A table starting at a large number can be linear; a table starting at 1 can be exponential with base 1.5. The starting value tells you the intercept, not the family. The differences and ratios tell you the family.

The two families also diverge in what the difference column does over time. In a linear table the differences are flat forever. In an exponential table the differences grow in proportion to the outputs, which is why real exponential data (bacteria counts, compound interest, viral spread) looks almost flat for several rows and then explodes β€” the differences stay small while the outputs are small. Scan the ratio column first if the outputs climb steeply; scan the difference column first if they look like a steady staircase.

A caveat on notation: some curricula write exponential functions as y = a Β· b^x and constrain b > 0, b β‰  1. If your ratios come out as 1 across the board, the function is technically exponential with base 1, which collapses to the constant function y = a. Most textbooks treat that as a degenerate case and call it constant rather than exponential. If a ratio comes out negative, the table cannot come from a standard exponential model β€” recheck the subtraction order before concluding anything exotic.

Three worked tables: one linear, one exponential, one that is neither

The cleanest way to internalize the method is to see all three outcomes. The three tables below use the same input column so you can compare directly.

Input Linear table Exponential table Neither (quadratic)
0 5 2 0
1 9 6 1
2 13 18 4
3 17 54 9

Linear column: differences are 4, 4, 4 and ratios are 1.8, 1.5, 1.3 β€” differences win, slope 4, intercept 5, so y = 4x + 5. Exponential column: differences are 4, 12, 36 and ratios are 3, 3, 3 β€” ratios win, base 3, start 2, so y = 2 Γ— 3^x. Quadratic column: differences are 1, 3, 5 and ratios are undefined at the zero start, then 4 and 2.25 β€” neither test produces a constant, and the second differences (3 βˆ’ 1 = 2 and 5 βˆ’ 3 = 2) are constant, which is the signature of a quadratic. Congratulations: you have also learned the test for polynomials one degree up.

Notice the exponential table's outputs at inputs 0 and 1 β€” 2 and 6 β€” look innocuous. A student eyeballing that table might guess linear because the early numbers climb modestly. By input 3 the output has reached 54 while the linear table sits at 17. Extrapolation is where the misclassification gets expensive, which is why the ratio test matters more than the visual impression.

How to handle a zero or negative output without abandoning the test

A zero output makes the ratio test fail for a mechanical reason: you can't divide by zero. That doesn't mean the data isn't exponential β€” it means the exponential model can't pass through that point, since a Β· b^x never equals zero for any finite x when a β‰  0. So a table containing an output of 0 is either linear, polynomial, or carries a shifted exponential that hasn't been transformed yet. Check differences first, and if they're constant, you're done.

Negative outputs are subtler. A standard exponential y = a Β· b^x can be negative if a is negative, which makes the entire output column negative, or if the input column includes negative x and a positive base β€” but the ratios of consecutive outputs will still be a positive constant b. If your ratio column alternates signs while the differences also break, the table likely represents an exponential multiplied by something oscillating, or it simply isn't a function family covered in your course.

Real-world tables hide boundary problems too. A population column rounded to whole people may show outputs 1000, 1050, 1103, 1158 β€” ratios 1.05, 1.0505, 1.0499 β€” which is constant enough to call exponential despite the rounding wobble. Set a tolerance before you compute: if your ratios agree to two decimal places, treat them as constant. A difference column that agrees to the nearest integer is similarly safe. Applied statistics routinely works this way, and a 2020 update to the U.S. Census Bureau's population estimates methodology explicitly addresses rounding of small-area counts β€” the ratios of those rounded figures can wobble by a few thousandths even when the underlying growth is smooth.

Reading real-world tables: population growth, depreciation, and savings

Every textbook example of exponential growth is a population table, and those tables carry a hidden trap. Suppose a town's population reads 8,000 in 2010, 8,400 in 2011, 8,820 in 2012 β€” ratios 1.05 and 1.05 exactly, so exponential at 5% per year. But a statewide table rounded to the nearest thousand might read 8,000, 8,000, 9,000, and the ratios become 1.0 and 1.125, which looks like blowup and isn't. Rounding destroys the ratio test in small counts. Ask for unrounded figures before ruling.

Car depreciation flips the sign of the story. A $30,000 vehicle worth $24,000 after one year, $19,200 after two, and $15,360 after three has ratios of 0.8 each year β€” an exponential decay function with base 0.8, also described as 20% annual depreciation. The differences (6,000, 4,800, 3,840) shrink as the car ages, which incidentally explains why depreciation hits hardest in year one and why a 20% rate never reaches zero; the model approaches it asymptotically.

Savings accounts with compounding interest behave identically on the growth side. A balance of 1,000, 1,050, 1,102.50, 1,157.63 has ratios of 1.05 across every row β€” 5% annual compounding, y = 1000 Γ— 1.05^t. The interesting comparison comes from putting compound interest beside simple interest in the same table: simple interest grows linearly at a fixed dollar amount (say $50 per year), while compound interest grows exponentially at a fixed percentage. At 5% on $1,000, both models pay roughly $50 in year one, but the compound model pays $52.50 in year two and $63.81 by year ten. That divergence between two tables with nearly identical first rows is exactly why the ratio test beats eyeballing.

Building the difference and ratio formulas in spreadsheets and calculators

Doing a dozen tables by hand is fine for class and miserable for homework. In a spreadsheet, put your inputs in column A and outputs in column B starting at row 2. In C3 type =B3-B2 and drag down for differences; in D3 type =B3/B2 and drag down for ratios. Format D as decimals to four places so the pattern is visible. The pattern jumps out: a linear column repeats the same value, an exponential column repeats the same multiplier.

Google Sheets and Excel both handle this identically, and both return an error in the ratio column if any output is zero β€” #DIV/0! in Sheets, #DIV/0! in Excel. That error is diagnostic, not fatal: a zero in the output column tells you the table isn't a plain exponential model, so fall back to the difference column. If you want to suppress the error for a clean printed sheet, wrap the ratio in =IFERROR(B3/B2,"β€”") and read the dash as "model can't pass through zero."

On a TI-84, enter inputs in L1 and outputs in L2 through the STAT edit menu, then run Ξ”List(L2) for differences and L2/Ξ”List(L1) paired with the previous-output shift for ratios; the calculator returns lists, so the constant-vs-changing pattern shows up immediately in the list editor. On a graphing calculator the regression route (ExpReg or LinReg) reports RΒ² for both models β€” and here's the trap: an exponential model can report a high RΒ² on perfectly linear data, because any smooth function fits a short stretch. Trust the difference and ratio columns, not RΒ².

Spotting the traps: tables that look exponential but are not, and a five-minute check

The most common false positive is a table whose inputs are not evenly spaced. Rows at x = 0, 0.5, 1, 1.5 with outputs 1, 2, 4, 8 look exponential β€” and they are, with ratio 2 per half-step, which is a base of 4 per unit step. But if the same outputs sit at x = 0, 1, 2, 3, the function is y = 2^x with base 2. The table alone can't tell you the base until you read the input spacing. Always confirm the step size before declaring the base.

A second false positive comes from proportional-looking triples. Outputs 2, 6, 18 have ratio 3 and are exponential. Outputs 2, 4, 6, 8 have constant differences and are linear. Outputs 2, 6, 10, 14 β€” differences of 4 β€” are linear even though the first two numbers also admit a ratio of 3 on one step. One ratio does not make an exponential sequence; you need the ratio to hold across at least two consecutive steps. This is the single most common wasted point on a quiz.

A third trap is the "starts at 1" assumption. Students often assume any table beginning with output 1 must be exponential because 1 = b^0. A linear table can start at 1 too: y = 1 + 3x begins at 1 and climbs 4, 7, 10. Starting values are intercepts, nothing more.

For a fast five-minute check on any table, work in this order: confirm the input step is constant; compute differences for three rows; if they match, stop and write the linear equation; if not, compute ratios for the same three; if the ratios match, stop and write the exponential equation; if neither matches, compute second differences to test for a quadratic. That order saves time because linear tables are more common in introductory exercises, and because a linear table's ratio column will look chaotic in a way that tempts you to abandon the ratio test too early.

From verdict to equation: turning the column you kept into slope or base

Once the differences are constant, the slope is that difference divided by the input step. For a table with step 1 and differences of 4, the slope is 4; the intercept is the output when the input is 0, which you can read directly if the table includes x = 0, or back-calculate as y βˆ’ mx if it doesn't. For the linear table 5, 9, 13, 17 above, the equation is y = 4x + 5.

Once the ratios are constant, the base is that ratio raised to the power 1 over the input step. A ratio of 3 per step of 1 gives base 3. A ratio of 4 per step of 2 gives base 4^(1/2) = 2 β€” a step most worksheets expect and many students miss. The starting value a is the output at input 0; if the table starts elsewhere, back-calculate with a = y Γ· b^x. For the exponential table 2, 6, 18, 54, the equation is y = 2 Γ— 3^x.

A worked example with a fractional step: outputs 10, 14.14, 20 at inputs 0, 1, 2 are ratios of 1.414 each β€” the square root of 2 β€” so the model is y = 10 Γ— (√2)^x. The ratios are equal to four decimal places when the outputs are rounded, which is your cue that the underlying function is exponential even though the printed numbers wobble. Rounding is not a disqualifier; a broken ratio pattern is.

The back-calculation matters because many tables begin at x = 1 rather than 0. Outputs 300, 900, 2700 at x = 1, 2, 3 give ratio 3 and imply a = 300 Γ· 3 = 100, so y = 100 Γ— 3^x. Students who read the first output as the starting value write y = 300 Γ— 3^x and get every subsequent value wrong by a factor of 3. This error is easy to catch in a grade check and hard to catch on a test.

When both tests fail: second differences and the polynomials hiding in your table

When neither the first differences nor the ratios are constant, the data isn't linear or exponential β€” but it usually is something with a name. Constant second differences identify a quadratic. The table 0, 1, 4, 9: first differences 1, 3, 5; second differences 2, 2. That's y = xΒ². The table 1, 4, 9, 16 behaves the same way. Once you can recognize the second-difference signature, you stop guessing and start classifying.

Higher-degree polynomials show the same pattern one level up. Constant third differences identify a cubic, constant fourth differences a quartic, and so on. This is the finite-difference method of polynomial interpolation, and it's the reason polynomial regressions can fit any sufficiently short table exactly β€” with n points, a polynomial of degree n βˆ’ 1 fits perfectly, which is precisely why the "fit a curve" button on a spreadsheet is not a classification tool.

Practical rule from all this: for introductory algebra, tables are linear, exponential, or quadratic in roughly that order of frequency, and the first-difference/ratio/second-difference sequence catches all three. If none of the three tests passes, the table may come from a rational function, a logarithm, or a piecewise model, and the honest answer is "none of the above." A quiz that asks "is this linear or exponential" will not hand you a table from outside those families, so if the tests fail, recheck your arithmetic before rechecking your worldview.

FAQ

Can a table be both linear and exponential?

Yes, in exactly one degenerate case: the constant function. Outputs 5, 5, 5 have differences of 0 and ratios of 1, so both tests pass. Most textbooks classify this as linear with slope 0 and exclude it from the exponential family by requiring a base other than 1. If your course follows that convention, answer "linear."

What if the table has only two rows?

Don't commit. Any two points lie on exactly one line and infinitely many exponential curves, so two rows are mathematically insufficient to decide the family. A fair question will give you the function's description or a third row. If forced, compute the difference and ratio, note that either could extend, and state the assumption.

Why does dividing by zero ruin the ratio test?

The ratio test divides consecutive outputs, and division by zero is undefined, so a table with an output of 0 yields an error in the ratio column. More fundamentally, no exponential function a Β· b^x can equal zero, so the table cannot come from the standard exponential model. Check the differences instead; if they're constant, the model is linear.

How do I tell linear from exponential if the inputs are unevenly spaced?

Convert to a per-unit rate first. Divide each first difference by the matching input step to get a per-unit slope, and take the nth root of each ratio (n = input step) to get a per-unit base. Constant adjusted values still indicate the same family; a table that jumps from x = 1 to x = 3 to x = 10 can be classified this way, but the raw columns will mislead you.

Is a table with constant ratios always exponential growth?

No β€” a constant ratio below 1 signals exponential decay. Ratios of 0.8 across the rows mean the outputs shrink 20% per step, so the model is y = a Γ— 0.8^x. The test doesn't distinguish growth from decay; the size of the base relative to 1 does. A ratio of exactly 1 returns us to the constant-function edge case.

Could an exponential table fool me with a high RΒ² from a linear regression?

Every time. A linear regression will report a strong RΒ² on a short stretch of exponential data because the first few points sit close to a straight line. The difference and ratio columns are the honest test; RΒ² measures fit to a chosen model, not the shape of the data, and it should never override a constant-ratio column.

Where do transformations like y = 2^x + 3 show up in tables?

A vertically shifted exponential breaks both tests in a specific way: its ratios are not constant but its differences grow by a nearly constant factor. Outputs 5, 7, 11, 19 (that's 2^x + 3) have differences 2, 4, 8 β€” which double each step β€” and second differences 2, 4, 8 as well, so the function is neither linear nor standard exponential. Introductory courses rarely test shifted families, but recognizing the pattern prevents a wrong answer.

Does it matter which row I subtract from which?

Yes. First difference means current output minus previous output, in the input's increasing order. Reversing the subtraction flips every sign and makes a linear table look like it has a constant negative difference β€” which is still constant, so the verdict survives, but the slope's sign flips. Fix one direction and stay consistent, or your slope will come out backwards.