What Makes a Prediction an Interpolation?
A fitted regression line can be used to predict a response for a chosen value of the explanatory variable. Before interpreting such a prediction, check where that explanatory value falls relative to the values in the data used to fit the line. A value within the observed range is an interpolation.
This check is about the predictor, not the predicted response. It does not tell us whether the prediction is close to a future observation, whether a linear relationship is a good description, or whether the explanatory variable causes the response to change. It answers a narrower question: is the predictor value inside the span of values used to fit the line?
The range must come from the explanatory-variable values in the data used to fit the particular regression line. Use the same variable and units as the fitted equation. For example, if the data used to fit a line run from 2 to 10 hours of training, a prediction for 7 hours is an interpolation. A prediction for 7 minutes is not directly comparable until its units have been converted to hours.
A Simple Check Before Calculating
For a fitted line written as \(\hat{y}=b_0+b_1x\), the predicted response at a chosen value \(x\) is found by substituting that value into the equation. To classify the prediction, first compare \(x\) with the observed minimum and maximum. The substitution gives the predicted response; it does not determine whether the prediction is an interpolation.
The classification itself requires no probability model or special assumption about the shape of the data. It requires an accurately identified range and a predictor value expressed on the same scale. If a fitted line is used to make a prediction, that line is the model for the calculation. Whether a straight line describes the relationship well is a separate question; being inside the observed range does not settle it.
Use the observations that were used to fit the regression line, not a different or larger data set.
Find the smallest and largest observed values of the explanatory variable, keeping their units.
If it is between the limits, including either endpoint, its prediction is an interpolation.
Substitute the predictor value into the fitted equation and report the predicted response in context.
Worked Example: Predicting Between Observed Training Times
Suppose a hypothetical data set records study time \(x\), in hours, and quiz score \(y\), in points, for four students. The observed study times are 2, 4, 6, and 8 hours, and the fitted regression line is \(\hat{y}=84-2x\). Classify and calculate the prediction for a student who studies for 5.5 hours.
The observed predictor range is from 2 to 8 hours. Since \(2\leq 5.5\leq 8\), this is an interpolation. The value 5.5 does not have to be one of the four recorded study times; it only has to lie within their range.
Substitute \(x=5.5\) into the fitted line:
The line predicts a quiz score of 73 points for a student who studies for 5.5 hours. The prediction is an interpolation because 5.5 hours lies inside the observed range, not because 73 is necessarily close to any student's actual score.
Why Values Between the Endpoints Count
The word interpolation reflects the position of the predictor value between the smallest and largest observed values. A value can be between those endpoints even if no observation has exactly that value. With measurements recorded at 1, 2, 3, and 4 hours, for instance, 2.5 hours is still inside the observed range. A regression line supplies a predicted response at 2.5 hours by evaluating the line there.
A short algebraic argument makes the endpoint idea precise. Let \(a=x_{\min}\) and \(b=x_{\max}\), with \(a<b\), and suppose \(a\leq x\leq b\). Define \(t=(x-a)/(b-a)\). Because \(x\) is between the endpoints, \(0\leq t\leq1\), and rearranging gives \(x=(1-t)a+tb\). Thus \(x\) is a weighted average of the endpoints. For the fitted line \(g(x)=b_0+b_1x\), substitution gives the same weighted-average relationship for its predictions:
Since the weights \(1-t\) and \(t\) are nonnegative and sum to 1, the line's predicted response at an in-range value is between its predictions at the two endpoints (or equal to an endpoint prediction). This is a property of a straight line. It does not prove that the line describes the underlying relationship accurately or that the observed responses follow a straight-line pattern.
Worked Example: A Predictor Value Not Recorded in the Data
In a hypothetical class exercise, the observed practice times are 1, 2, 3, and 4 hours. A fitted line for a test score, measured in points, is \(\hat{y}=48+3x\). Classify and calculate the prediction at 2.5 hours.
The observed range is 1 to 4 hours. The proposed value satisfies \(1\leq2.5\leq4\), so this prediction is an interpolation, even though 2.5 hours is not among the recorded practice times.
The predicted score is:
The line predicts 55.5 points at 2.5 hours of practice. This number is the output of the fitted equation, not a guarantee that a student practicing for that long would earn exactly 55.5 points.
The endpoint predictions are \(\hat{y}=48+3(1)=51\) points and \(\hat{y}=48+3(4)=60\) points. The prediction of 55.5 points lies between them, as the weighted-average argument predicts.
Check the Data Set and the Units
A range check is only meaningful if it refers to the observations that produced the line. If a software output reports a regression equation fitted to one subset of records, use the predictor range in that subset. Do not silently use the range from a different group, a later data collection, or a combined file. Those changes alter the reference range and may alter the fitted line as well.
Units matter for the same reason. Suppose the data range is 30 to 90 minutes, but the proposed predictor value is stated as 1 hour. Convert 1 hour to 60 minutes before comparing it with the range. Conversely, if the equation uses hours, enter 1 hour rather than 60. A correct numerical comparison in mismatched units can lead to the wrong classification.
Worked Example: Comparing Predictions With the Correct Range
A hypothetical data set relates weekly exercise time to a fitness measure. The observations used to fit a line have exercise times of 5, 8, 12, and 15 hours per week. The fitted equation is \(\hat{y}=40+1.6x\), where \(\hat{y}\) is measured in fitness-score points and \(x\) is hours per week. Classify predictions at 6 hours and 18 hours, and calculate the prediction at 6 hours.
The observed range is 5 to 15 hours per week. Since \(5\leq6\leq15\), the prediction at 6 hours is an interpolation. Since 18 hours is greater than the observed maximum of 15 hours, the prediction at 18 hours is outside the observed range. This comparison classifies the predictor values; it does not by itself assess the quality of either numerical prediction.
For 6 hours, substitute \(x=6\):
The fitted line predicts 49.6 fitness-score points at 6 hours of exercise per week. The interpolation classification follows from 6 being inside the 5-to-15-hour range.
What Interpolation Does—and Does Not—Tell You
Interpolation is useful because it identifies predictions made within the span of the predictor values represented in the fitting data. It prevents a common category error: describing every prediction at a value not recorded exactly as outside the data's range. If observations span 5 to 15 hours, a prediction at 9 hours is an interpolation whether or not any observation occurred at exactly 9 hours.
However, “interpolation” is not a stamp of accuracy. A fitted line can be a poor summary even over the observed range, for example if the relationship is curved or if observations vary substantially around the line. The range check also says nothing about how precisely the response can be predicted. It only locates the chosen explanatory value relative to the observed minimum and maximum.
A useful report separates the calculation from its classification: give the predictor value and its units, state that it lies within the observed range, then report the fitted prediction and its response units. If relevant, note that the value is an estimate from the line rather than a guaranteed outcome.
Common Mistake
A frequent mistake is to say, “There was no observation at 2.5 hours, so a prediction at 2.5 hours is outside the data and cannot be an interpolation.” This confuses an exact recorded value with the full range of recorded values. If the observed predictor values run from 1 to 4 hours, then \(1\leq2.5\leq4\); 2.5 is within the range and the prediction is an interpolation. The correct check uses the minimum and maximum, not whether the proposed value appears as a particular row in the data table.
Check Your Understanding
For each question, focus first on the observed range of the explanatory variable.
- A line was fitted using temperatures from 10°C to 24°C. Is a prediction at 18°C an interpolation? Explain.
- Observed values of \(x\) are 3, 5, 8, and 11. A proposed prediction uses \(x=7\). Does 7 need to be an observed value for this to be interpolation?
- A data set used to fit a line has predictor values from 20 to 70 minutes. A proposed value is 1 hour. What conversion is needed before classifying it?
- Explain why being an interpolation does not establish that a fitted regression line predicts accurately.
- A fitted line is \(\hat{y}=12+4x\), and the fitting data have \(x\)-values from 2 to 6. Calculate and classify the prediction at \(x=5\).