Prediction Rounding and Units

From a Calculated Value to a Clear Prediction

In “Predictions From Computer Output,” you practiced reading coefficient estimates, writing \(\hat{y}=a+bx\), and substituting a predictor value. A calculator may display many digits for the result. Reporting all of them can suggest more precision than the context supports, while rounding too early can change the answer. This tutorial focuses on the last step: choosing a sensible rounded value, attaching the response units, and explaining what the prediction means.

The fitted line gives a numerical prediction for the response. That number should be reported with the units of the response variable—not the predictor—and in a sentence that says what predictor value was used. Rounding does not make the prediction exact, nor does it guarantee that an individual response will equal the predicted value.

Choose a Sensible Rounding Precision

First, check whether the question specifies how to round. If it says “to the nearest tenth,” “to the nearest whole unit,” or “to the nearest cent,” follow that instruction. If it gives no rounding rule, use a precision that is useful in the setting and consistent with the scale on which the response is measured. For example, reporting a predicted travel time as 18.426731 minutes is rarely useful; reporting it as about 18.4 minutes may be clearer.

There is no single number of decimal places that is correct for every regression prediction. The appropriate choice depends on the response and the purpose of the prediction. Money is often reported to the nearest cent when that level of detail is relevant. A measurement recorded to the nearest whole centimeter may call for a whole-centimeter report. A time in minutes may reasonably be reported to a tenth of a minute, or as a whole number of minutes if that is the requested precision.

Keep the unrounded calculator result while working, then round only the final value. Intermediate rounding can introduce avoidable error, especially when the slope has several decimal places or the predictor value is large. If a calculator returns 7.1132 and the question asks for the nearest tenth, report 7.1—not a result calculated by first rounding the slope or product.

Rounding is about how the answer is written; it does not change the meaning of the fitted line. In particular, do not confuse a rounded prediction with a claim that the observed response is known to that exact value. The line predicts an average pattern for the response at the stated predictor value; an individual response may be higher or lower.

Keep Units Attached to the Right Quantities

The response \(y\) has response units, and the predicted response \(\hat{y}\) has those same units. The intercept \(a\) is measured in response units. The slope \(b\) has response units per predictor unit, as explained in “Slope Units and Rates of Change.” Multiplying the slope by a predictor value gives a quantity in response units, which can be added to the intercept.

This units check can reveal a mistake. If a line predicts minutes from hours of practice, the prediction is in minutes, even though the input is in hours. Do not report the prediction in the predictor’s units. And do not attach the slope’s “per predictor unit” units to the final predicted response: those belong to the rate of change, not to \(\hat{y}\).

Pay attention to how the predictor is recorded. If the model uses distance in kilometers, substitute a value in kilometers. If the question gives miles instead, convert the input to kilometers before evaluating the line, or use a correctly converted model. The same principle applies when a predictor is recorded in hundreds, thousands, or another scaled unit: use the scale specified for the equation.

Worked Example: Reporting a Battery-Life Prediction

A fictional technology class models battery life using screen brightness. Let \(x\) be brightness in percentage points and \(y\) be battery life in hours. The fitted line is \(\hat{y}=9.482-0.0376x\). Predict battery life at a brightness setting of 63 percentage points, and report the answer to the nearest tenth of an hour.

The final number has hours as its units because battery life is the response. The input, 63 percentage points of brightness, is named in the sentence so the prediction is clear. Reporting 7.1132 hours would ignore the requested rounding; reporting 7.1 percentage points would give the predicted value the wrong units.

Worked Example: Rounding a Cost Prediction

Suppose a fictional scooter-rental class fits a line to predict a ride’s cost from its duration. Let \(x\) be ride duration in hours and \(y\) be cost in dollars. The fitted line is \(\hat{y}=2.50+1.85x\). Find and report the predicted cost for a ride lasting 3.75 hours, to the nearest cent.

This example also shows why the product should not be rounded prematurely. Rounding \(6.9375\) to \(6.94\) before adding happens to give the same final cent here, but that is not a reliable general strategy. Carry the available precision through the calculation, then round once at the end.

Write the Prediction in Context

A complete prediction sentence tells the reader what the input was and what response the line predicts. A useful pattern is: “For [predictor value and units], the fitted line predicts [rounded response and units].” Use “predicts,” “estimated,” or “about” when those words help make clear that the number is a model-based prediction rather than a guaranteed observation.

Avoid a bare number such as “10.0,” because the reader cannot tell what it measures or which predictor value produced it. Avoid saying “the response is exactly 10.0,” because the fitted line supplies a prediction, not certainty about the actual response. As covered in “Prediction Versus Observed Values,” the observed response for an individual can differ from its predicted response.

Worked Example: Predicting a Wait at a Community Event

A fictional community-event team models the average wait at an entrance using the number of visitors arriving in a short period. Let \(x\) be arrivals in thousands of visitors and \(y\) be wait time in minutes. The fitted line is \(\hat{y}=4.73+0.82x\). Predict the wait when 6.4 thousand visitors arrive, and report the result to the nearest tenth of a minute.

The scale used for \(x\) matters. Substituting 6,400 would treat the model as though its predictor were individual visitors, producing a very different result. Naming “6.4 thousand visitors” in the conclusion makes the scale explicit and ties the prediction to the situation.

A Quick Check Before You Report

Before writing a final prediction, read the question once more and check the input, output, rounding, and sentence. The sequence below is useful whether you evaluated the line by hand or used a calculator.

Common Mistakes and AP Exam Tips

• Reporting every calculator digit. A display with many digits does not mean the prediction is that precise. Follow the question’s rounding instruction; otherwise, choose a useful precision for the context.
• Rounding too early. Rounding the slope or product before finishing can alter the result. Keep full available calculator precision and round the final prediction once.
• Using predictor units on the answer. The prediction is \(\hat{y}\), so it has the response units. Check the variable roles before attaching units.
• Ignoring a scaled predictor. If \(x\) is measured in thousands, enter the number of thousands. Do not enter the unscaled count unless the equation uses individual units.
• Giving only a number. A full-credit response identifies the predictor value and predicted response in context, with response units and sensible rounding.
• Calling a prediction exact or guaranteed. Say that the line predicts or estimates the response. Do not claim that a particular individual’s observed response must equal the prediction.

A strong AP-style response makes the calculation traceable and the conclusion meaningful: show the substitution, give a sensibly rounded result, attach the response units, and state what predictor value it corresponds to. The result is a prediction from the fitted line, not a promise about an individual outcome.