Residual analysis examines the differences between observed values and values predicted by a model. The core aim of Science mastery is not to make students calculate leftovers after regression. It is to help them ask whether the model captured the scientific pattern—or whether structure is still hiding in the errors.
For students and parents searching for residual analysis, residual plot, regression residuals, model fit or how to interpret residual plots, the key principle is simple: a suitable model should leave residuals that look like unexplained scatter rather than another visible pattern.
If residuals still tell a story, the model has not finished telling it.
The 60-Second Formula
Residual = observed value − predicted value.
- Positive residual: observation is above prediction.
- Negative residual: observation is below prediction.
- Zero residual: observation matches prediction.
Wait, What? A Strong Fit Statistic Can Still Hide a Bad Pattern?
Yes. Summary statistics can look impressive while residuals reveal curvature, changing spread, clusters or outliers. This is why model diagnostics matter.
A Worked Example
Observed value = 18. Predicted value = 15.
Residual = 18 − 15 = +3.
The model underpredicted the observation by 3 units.
Residual Plots
A residual plot usually places fitted or predictor values on the horizontal axis and residuals on the vertical axis. A zero line makes systematic patterns easier to see.
For a suitable linear model, residuals should generally scatter around zero without a strong pattern.
Curved Residual Patterns
If residuals form a curve, a straight-line model may be missing a nonlinear relationship.
Funnel-Shaped Residuals
If residual spread widens as fitted values increase, variability may not be constant. At more advanced levels this is called heteroscedasticity.
Outliers
A large residual may indicate measurement error, a rare observation, a different subgroup or an incomplete model. Investigate before deleting.
Residuals and Linear Regression
Regression fits the model. Residual analysis checks whether the fitted model remains scientifically defensible.
See Linear Regression.
Residuals and Prediction
A model can capture an average trend while predicting individual observations poorly. Residual size makes that limitation visible.
Residuals Are Not Simply Mistakes
Residual variation can reflect measurement error, natural variation, unmeasured variables and model simplification. A residual is the part the model did not predict.
How to Practise
- Calculate predicted values.
- Calculate residuals.
- Plot them.
- Look for structure.
- Decide whether the model needs revision.
Common Mistakes
- reversing observed and predicted values;
- assuming residuals are experimental mistakes;
- looking only at R²;
- ignoring curvature;
- deleting large residuals automatically.
Frequently Asked Questions
What is a residual?
Observed value minus predicted value.
What should a good residual plot look like?
For a suitable linear model, residuals should generally scatter around zero without a systematic pattern.
What does a curved residual plot suggest?
The straight-line model may be missing nonlinearity.
Useful eduKateSG Routes
The Core Aim
Residuals show what the model failed to predict.
Calculate the difference. Plot it. Look for structure. Improve the model when the errors are trying to tell you something.
That is the core aim: judge a model by what remains after the fit, not only by how impressive the fitted line looks.
Properly taught kids shine a bright light into the future.