When Cross Multiplication Gives You the Wrong Answer

A 20 mile ride costs $43.00 when the fare is a $3.00 pickup fee plus $2.00 a mile. Build a proportion from the 5 mile fare and it predicts $52.00 instead. The arithmetic is neat, internally consistent, checkable in both directions, and wrong by $9.00.

Cross multiplication is not the problem. The assumption underneath it is: a proportion describes a straight line through the origin. Every fixed cost, threshold and limit breaks that assumption, and the arithmetic cannot detect the mismatch. It returns a clean number for a relationship that never supported one.

The method is fine; the assumption underneath it is what fails

On the proportion calculator, 3/4 = 9/x gives x = 12, with both cross products landing on 36. The recipe case behaves the same way: 4/300 = 7/x gives 525 g of rice. In each of those the two ratios genuinely are equal, so solving the proportion answers the question that was asked.

The danger sits one step earlier, at the moment you decide two ratios ought to be equal. Cross multiplication solves an equation you have already written down. It never checks whether the quantities in it actually vary together in the way the equation claims, and flawless algebra on a wrong model returns an exact answer to the wrong question.

A proportional rule has the form y = mx. Its graph is a straight line, its rate m stays constant across the whole range, and its value at x = 0 is zero. Add a fixed charge and the rule becomes y = mx + b, which shifts the entire line up and away from the origin by b while leaving the slope untouched.

That shift is the whole story. The ratio calculator will happily compare the two numbers in any observed pair, and the ratio it returns is correct. It becomes a scaling rule only when the relationship behind the pair passes through the origin, and nothing in the pair itself tells you whether it does.

The one question: does zero in give zero out?

Before setting up a proportion, ask whether zero input would produce zero output. If it would not, stop. You have found a pickup fee, a base charge, a minimum order, a handling cost or some other offset, and a proportional model has no way to represent it.

The ride fails on sight. Zero miles still costs $3.00, because the pickup fee is charged before the wheels turn. That puts the point (0, $3.00) on the fare line and leaves the origin off it. Any proportion built from a non-zero trip quietly spreads that $3.00 across the miles as though it were part of the per-mile rate.

A real conversion passes. One mile is exactly 1.609344 km, so 10 miles is 16.09344 km, and zero miles is zero kilometres with nothing left over. The relationship behind the miles to kilometres converter scales cleanly because it has no starting amount to protect from the multiplication.

Passing the zero test is necessary rather than sufficient. A curved relationship can run through the origin without being proportional, and a threshold or a cap can change the rule partway along the range. So after checking zero, ask the second question: does one constant rate govern the whole span you are scaling across?

The worked failure: a fare with a pickup fee

Take a ride priced at a $3.00 pickup fee plus $2.00 a mile. These figures are chosen to make the arithmetic legible and are not any real operator's pricing. The correct model is cost = 2 × miles + 3, and every number below comes out of it.

At 5 miles the true cost is $13.00. Treating that pair as proportional makes the apparent unit rate 13 divided by 5, but that quotient has the pickup fee folded into it. Scaling the quotient assumes every part of the $13.00 grows with distance, when $3.00 of it should be paid exactly once no matter how far the ride goes.

The two columns agree at 5 miles because that is the row the proportion was built from. Agreement there proves nothing about any other row, and it is exactly the row a person checks when they want reassurance. Everywhere else the prediction drifts: $26.00 against $23.00, $39.00 against $33.00, $52.00 against $43.00.

Notice that the error also changes direction below the reference point. At 2.5 miles the proportion predicts $6.50 against a true $8.00, understating the fare by $1.50. Scaling up charges the pickup fee too many times, and scaling down refunds part of a fee that was never divisible in the first place.

The error law: (k − 1) times the fixed part

The mistake is not only detectable, its size follows a formula. Call the known input x, the rate m and the fixed part b. The true output at the known point is mx + b, and the true output after scaling the input by k is mkx + b, because the fixed part does not scale.

A proportion instead multiplies the whole known output by k, predicting (mx + b)k. Subtract the true value from the prediction and the variable terms cancel, since both contain m times k times x. What survives is (k − 1) × b, the fixed part counted one extra time for every multiple of the reference point.

So doubling gives k = 2 and an error of $3.00, which is precisely one pickup fee. Tripling costs $6.00 and quadrupling $9.00, each figure landing on a whole number of pickup fees rather than on anything that needs computing. The law does the work that a worked example only illustrates.

Scaling down reverses the sign without needing a separate rule. At half the reference distance k is below one, so (k − 1) is negative and the predicted $6.50 falls $1.50 short of the true $8.00. One formula covers both directions, which is a good sign that it is describing the structure rather than a symptom.

The same formula names the condition under which cross multiplication is safe. When b is zero the error is zero at every value of k, without exception. That is the algebraic form of the statement that a proportion is a line through the origin, read from the other side and turned into something you can test.

The absolute error grows in fixed steps, while the relative error climbs towards a ceiling. The prediction runs 13.0% high at double, 18.2% at triple and 20.9% at quadruple. It never passes 30%, which is the pickup fee divided by the variable part of the reference fare, and that ratio is the share of the original the proportion has misfiled as rate.

The repair: two points recover both parts

One observed point cannot tell y = mx from y = mx + b. An unlimited number of straight lines pass through any single point, including the proportional one forced through the origin and every offset line with a shallower slope. Nothing you do to one measurement will reveal a fixed part, because the measurement does not contain that information.

Take (5 miles, $13.00) and (10 miles, $23.00). The output rises by $10.00 while the input rises by 5 miles, so the rate is $2.00 a mile. That rate is the slope of the line through the two points, and it is the same quantity the slope calculator returns from any two coordinates you hand it.

With the rate known, either point recovers the rest. At 5 miles the variable portion is $10.00 against a total of $13.00, leaving $3.00 as the fixed part. The model is now cost = 2 × miles + 3, and putting 20 into it returns $40.00 plus $3.00, which is $43.00 exactly.

The repair works because it keeps the two components apart instead of averaging them into a single rate. It has one condition of its own: two points identify an offset line only where a constant rate applies. If a threshold changes the rate partway, points taken from either side of it will hide the change rather than reveal it.

The other family: relationships that are not straight at all

Some proportions fail with no fixed part anywhere in sight. They fail because the relationship is curved, or because it follows a different power of the scale factor. Doubling a room's side lengths gives 4 times the floor area and 8 times the volume, so neither of those quantities doubles when the length does.

That is a different diagnosis and it needs a different repair. There is no b to subtract, and the (k − 1) law does not describe it, so removing a fixed part would fix nothing. The scale-factor guide covers that family: lengths take k, areas take k squared, volumes take k cubed.

Percentage changes compose rather than add, which is the same trap in a third costume. A 20% discount followed by a 10% discount leaves 72% of the price, so 28% came off rather than 30%, and the percentage calculator works the compounding out properly.

The zero test stays useful across all three families as a first filter rather than a certificate. Fail it and proportionality is already ruled out. Pass it and you still have to ask whether one constant rate runs the length of the range before you let equal ratios stand in for a model.

Common mistakes that produce convincing answers

• Treating one pair as proof of proportionality. A known pair gives you a ratio, not evidence that the ratio holds anywhere else. Run the zero test, then take a second measurement whenever a fixed component is plausible.

• Calling the total-per-unit quotient the rate. Dividing $13.00 by 5 miles hands part of the $3.00 pickup fee to every mile. That quotient is an average cost per mile and it changes with distance; the actual rate is a flat $2.00.

• Checking only the reference point. A proportion must reproduce the fare it was built from, so that check can never fail and therefore never tells you anything. Test a different input, and test zero first of all.

• Assuming that scaling down is the safe direction. Halving the input does not halve a fixed part, so the error simply flips sign: $6.50 predicted against $8.00 true. The negative value of (k − 1) says so before you compute anything.

• Reading a modest percentage error as validation. The fare is 13.0% high at double and 20.9% high at quadruple. Both come from one false assumption, and neither figure is small enough to ignore or large enough to notice by feel.

• Flattening a threshold or a limit into one rate. A rule that changes after a threshold, or stops changing at a cap, is piecewise across its range. Apply the rule that governs each region rather than fitting one line through all of them.

• Using the fixed-part law on a curved relationship. The (k − 1) formula diagnoses an offset straight line and nothing else. Areas, volumes and stacked percentages each need their own scaling rule, not a proportion with a correction bolted on.

A proportion assumes the line passes through the origin. Ask whether zero in gives zero out, then confirm that one constant rate covers the range you are scaling across. If a fixed part exists, scaling by k makes the answer wrong by exactly (k − 1) times that part, however clean the cross multiplication looks on the page.