Residuals and model fit
A residual is observed output minus predicted output. Examine patterns as well as size: a curved residual pattern may indicate a linear model misses structure, even when some individual errors are small. Work through the example in two forms, such as an annotated calculation and a labeled diagram, table or verbal explanation. Name the assumption that makes the operation valid, and compare a tempting incorrect approach with the justified method. Keep an exact value until the problem asks for rounding; a final explanation should connect the result to the quantities being compared.
A worked example
A model predicts 10; the observed value is 12. What is the residual observed−predicted? Worked solution: Residual=12−10=2. A residual records a prediction error; it does not by itself establish a cause. The result can be checked against the stated quantities and units before it is used in a recommendation.
- Explore residuals and model fit using the worked scenario. Sketch or tabulate the quantities, label the unknown, and predict the kind of answer before calculating.
- Resolve this new case without copying the model: A model predicts 13; the observed value is 16. What is the residual observed−predicted? Show each equality or representation change and explain its purpose to a partner or in writing.
- Complete the two practice questions about residuals and model fit. For one tempting wrong choice, describe the likely reasoning error and construct a small counterexample.
- Create and solve an original residuals and model fit problem with new quantities. State its assumptions, include a representation, and check that the requested answer has appropriate units.
- Answer the independent check questions without viewing the worked solution. Then compare methods, revise one explanation, and record which idea needs another example.
Open the printable learning material