Breusch–Pagan Test & Heteroscedasticity

Ordinary least squares can estimate a mean relationship even when residual spread changes, but familiar standard-error formulas rely on a variance model. The Breusch–Pagan approach asks whether squared residuals are systematically related to selected explanatory variables. An auxiliary regression summarises that relationship, and its R-squared becomes a Lagrange-multiplier statistic after multiplication by sample size. This is why mathematics matters in regression diagnostics: a visual fan shape becomes a testable variance question, while the limits of that test remain explicit. Preserve the evidence trail: state the mathematical object, calculate a small case, verify the result independently, vary one assumption and explain the boundary. That predict–calculate–check–critique cycle builds numeracy and transferable problem-solving without promising intelligence, admission, income or employment from one method. For a fuller investigation, keep a computation log with the original input, every declared convention, the hand-worked result, the software result, a residual or reconstruction check, and one deliberately awkward case. Ask what information the method uses, what it discards, which assumption provides the guarantee, and which quantity measures failure. Repeat after changing scale, precision or one parameter. A student should explain why the output moved, not merely observe that it changed. Parents and educators can support this by valuing clear diagrams, labelled units, honest uncertainty and corrected reasoning as much as speed.

Quick navigation: Why this mathematics matters · How the mechanism works · Worked example and core equations · Four deeper ideas · Twelve practical investigations · Limits and misconceptions · A learning route for students · Guidance for parents and educators · Frequently asked questions · Useful next reading and sources

Why this mathematics matters

Mathematics is important here because the useful result is not merely a picture or a fast answer. It is a chain of reasons: a representation records selected information; an equation transforms it; an invariant explains what survives; and a limit states where the conclusion may fail. Students who can narrate that chain are learning transferable problem-solving skills, not only specialist vocabulary.

The search intent behind ‘why mathematics is important’ is often practical: where do algebra, geometry, probability and numerical reasoning actually earn their place? This mechanism gives a concrete answer. It turns a large task into smaller checkable operations, and it lets a learner distinguish a theorem, an implementation convention, a measured result and an interpretation.

Did You Know? The foundational source is Breusch and Pagan's 1979 Econometrica paper. A current implementation reference is statsmodels' official het_breuschpagan documentation. Reading both prevents a common mistake: attributing every modern feature to the original publication or assuming software defaults are universal mathematics.

How the mechanism works

First fit the intended regression and retain its residuals. Square those residuals so larger absolute deviations receive larger positive values. Choose variables that might explain the error variance, normally including a constant, and regress the squared residuals on them. Under the classic formulation and null hypothesis of constant variance, a scaled measure of explained variation has an asymptotic chi-square distribution. The commonly used Koenker version relaxes the normal-error assumption while retaining independent and identically distributed errors under the null. Modern software may report both an LM statistic and an F-statistic, so the selected variant and degrees of freedom must be recorded.

In the familiar auxiliary-regression form, LM=nR-squared, where n is the number of observations and R-squared comes from regressing squared OLS residuals on q variance covariates plus a constant. Under the null and regularity conditions, LM is asymptotically chi-square with q degrees of freedom. The F form tests the same auxiliary regression through its explained and unexplained sums of squares. The original Breusch–Pagan derivation assumes normally distributed disturbances; the studentised Koenker variant is robust to non-normality but still depends on appropriate independence and model structure. The test detects association with the supplied variance regressors, not every possible heteroscedastic pattern.

A useful four-column notebook for this topic is object, rule, guarantee, caveat. In the first column, name what the algorithm receives and stores. In the second, write one operation without skipping its conditions. In the third, state exactly what the mathematics guarantees. In the fourth, record a boundary case, approximation or modelling choice. This small routine makes advanced mathematics readable because it keeps symbols attached to meaning.

Worked example and core equations

Suppose six residuals are [-1,1,-2,2,-3,3] and a suspected scale variable is z=[1,1,2,2,3,3]. The squared residuals are [1,1,4,4,9,9]. Regressing them on a constant and z gives fitted values 2/3,2/3,14/3,14/3,26/3,26/3. The residual sum of squares is 4/3 and the total sum of squares is 196/3, so R-squared=1-(4/196)=48/49 about 0.9796. Therefore LM=6×48/49=288/49 about 5.8776. Compared with chi-square on one degree of freedom, the asymptotic p-value is about 0.0153. The arithmetic shows the mechanism clearly, but six observations are far too few to trust the asymptotic calibration without caution.

How to audit the calculation

Recompute each intermediate value from the definition before relying on a shortcut.
Keep indices, coordinate order, units and normalisation visible.
Test one case with a known exact answer.
Name any random choice, tolerance or boundary convention.
Compare with a slower reference calculation whenever possible.
Separate numerical agreement from proof that the real-world model is appropriate.

The worked numbers are deliberately small. A hand calculation can expose a sign error, reversed convention or missing scale factor before thousands of values make the same error harder to see. After the tiny case succeeds, increase size gradually and measure both accuracy and computational work.

Four deeper ideas

Squaring residuals converts spread into a response

Ordinary residuals have signs and often sum to zero with an intercept, so a simple regression of residuals on predictors would study the conditional mean again. Squaring removes sign and makes magnitude visible. It is an imperfect proxy for conditional variance because each squared residual is noisy, but across observations a systematic pattern can reveal a variance relationship.

The auxiliary regression turns a nuisance pattern into explained variation

R-squared asks how much variation in squared residuals is aligned with the proposed variance variables. Multiplying by n produces the score-style LM statistic. This is a recurring statistical strategy: fit the simpler null model, study how its residual evidence points toward a richer alternative, and measure the direction without fully trusting a plot.

A rejection diagnoses, but does not prescribe one remedy

Heteroscedasticity can arise from scale effects, omitted nonlinear terms, mixtures of groups, measurement error or a genuinely changing conditional variance. Robust standard errors may improve inference for coefficients without modelling the variance; weighted least squares can improve efficiency when weights are justified; transformations may change interpretation. The cause and purpose determine the response.

Mean misspecification can imitate variance misspecification

If the fitted mean omits curvature or interactions, residuals can form a structured pattern whose magnitude changes with a predictor. A Breusch–Pagan rejection may then be a clue that the mean model is incomplete. Residual-versus-fitted plots and substantive knowledge should accompany the test rather than treating the p-value as a machine that labels one isolated assumption.

Twelve practical investigations

1. Income And Spending

Residual spread grows at higher household income. Regress squared residuals on income and inspect nonlinear alternatives.

What the mathematics reveals: Scale can be represented as a variance regressor. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: Income measurement and omitted household structure may drive the pattern. This boundary belongs beside the result so the example remains useful rather than overstated.

2. School Score Model

Predicted scores near the middle have wider residuals than those near boundaries. Include fitted value and a squared fitted-value term in a broader diagnostic strategy.

What the mathematics reveals: A linear-only auxiliary regression may miss curved variance. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: Bounded outcomes can create mechanical variance changes. This boundary belongs beside the result so the example remains useful rather than overstated.

3. Hand Calculation

Six paired residual magnitudes increase with z. Compute the auxiliary fitted values, sums of squares, R-squared and nR-squared.

What the mathematics reveals: The LM statistic comes from ordinary regression geometry. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: Asymptotic p-values are weak evidence at n=6. This boundary belongs beside the result so the example remains useful rather than overstated.

4. Classic Versus Koenker

Software offers robust=True by default. Record whether the original normality-based or Koenker studentised variant is used.

What the mathematics reveals: A familiar test name can hide different assumptions. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: Do not report a method label without its variant. This boundary belongs beside the result so the example remains useful rather than overstated.

5. Robust Inference

The coefficient estimand remains meaningful despite changing variance. Report heteroscedasticity-consistent standard errors with a justified finite-sample choice.

What the mathematics reveals: Inference can be adjusted without changing point estimates. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: Robust does not mean valid under dependence or endogeneity. This boundary belongs beside the result so the example remains useful rather than overstated.

6. Weighted Least Squares

Measurement precision is known from instrument calibration. Use defensible inverse-variance weights and compare residual diagnostics.

What the mathematics reveals: A variance model can improve efficiency. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: Estimated or arbitrary weights may introduce new errors. This boundary belongs beside the result so the example remains useful rather than overstated.

7. Omitted Curvature

Residuals form a U-shape against x. Repair the mean specification before interpreting a variance test.

What the mathematics reveals: Squared residual structure can originate in mean bias. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: Treating all structure as heteroscedasticity misses the first problem. This boundary belongs beside the result so the example remains useful rather than overstated.

8. Time Series

Residual spread and autocorrelation both appear over time. Use diagnostics and covariance methods designed for serial dependence.

What the mathematics reveals: Variance and correlation are different second-moment features. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: A basic Breusch–Pagan test does not solve autocorrelation. This boundary belongs beside the result so the example remains useful rather than overstated.

9. Clustered Classrooms

Students are nested within classes. Use a multilevel or cluster-aware analysis and inspect within-cluster variance.

What the mathematics reveals: Shared contexts violate independent-observation calculations. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: More students in one class do not equal independent replications. This boundary belongs beside the result so the example remains useful rather than overstated.

10. Many Candidate Regressors

A researcher tries twenty separate variance tests. Predeclare a small diagnostic set or adjust the exploratory interpretation.

What the mathematics reveals: Repeated testing changes the false-positive story. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: Selecting only the smallest p-value exaggerates evidence. This boundary belongs beside the result so the example remains useful rather than overstated.

11. Simulation Power

A mild variance slope may be hard to detect. Simulate from the fitted design at several effect sizes and sample sizes.

What the mathematics reveals: Power depends on design and alternative shape. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: Simulation inherits every assumption coded into it. This boundary belongs beside the result so the example remains useful rather than overstated.

12. Reporting A Remedy

The diagnostic rejects constant variance. State the estimand, suspected cause, selected response and post-remedy checks.

What the mathematics reveals: A diagnostic is part of a reasoning chain, not the final paragraph. Ask the learner to show the precise equation, inequality, index rule or invariant that supports this conclusion, then check it on one small example.

Limit to keep visible: Do not claim causation from residual association. This boundary belongs beside the result so the example remains useful rather than overstated.

Limits and misconceptions

The Breusch–Pagan family relies on an adequately specified mean model and a chosen set of variance regressors. It may miss nonlinear variance patterns not represented by those variables. The classic version is sensitive to non-normal disturbances; the Koenker version changes assumptions but does not handle clustering, serial correlation or arbitrary dependence automatically. Small and moderate samples can make the chi-square approximation liberal, and statsmodels notes that an F-statistic may be preferable in such settings. Multiple diagnostic searches inflate false-positive risk. A rejection does not identify a causal mechanism, while non-rejection does not prove constant variance.

Five misconceptions to challenge

Heteroscedasticity makes OLS coefficients automatically biased: Under exogeneity the coefficients can remain unbiased, while efficiency and usual standard errors are affected.

One significant test reveals the cause: It detects an association with selected variance regressors, not a causal explanation.

Robust standard errors remove heteroscedasticity: They adjust an inference calculation; the residual variance pattern still exists.

A large p-value proves homoscedasticity: It may reflect limited power, poor variance regressors or an inadequate sample.

Every fan-shaped plot requires a response transformation: Remedies depend on the estimand, mean model, variance process and communication needs.

Another broad misconception is that advanced mathematics automatically creates intelligence, admission, income or employment. It does not. Learning it can strengthen modelling, abstraction, calculation and explanation when practice is deliberate, but opportunities also depend on interests, communication, domain knowledge, education pathways and many circumstances outside one topic. Keep claims specific and options open.

A learning route for students

Move from a visible trace to a symbolic rule, then to code and critique. For every step below, prepare three pieces of evidence: a correct hand example, a boundary or failure case, and a short explanation using the relevant invariant. Do not advance merely because a library returned a number.

Step 1: Calculate and plot regression residuals

Start with the smallest non-trivial instance. Predict the result, perform the calculation, and compare it with a reference. Change one assumption—size, spacing, sign, threshold, tolerance or data distribution—and explain which part of the reasoning changes. Finally, teach the step without code. That predict–calculate–vary–explain cycle is a strong test of transferable understanding.

Step 2: Square residuals and retain observation alignment

Step 3: Fit an auxiliary regression

Step 4: Compute r-squared from sums of squares

Step 5: Calculate the nr-squared lm statistic

Step 6: Count degrees of freedom correctly

Step 7: Distinguish classic and koenker variants

Step 8: Separate mean and variance misspecification

Step 9: Choose between robust inference and variance modelling

Step 10: Report diagnostic uncertainty without causal overclaiming

A four-week practice plan

Week 1 — vocabulary and diagrams: define each object, redraw the core mechanism and reproduce the worked example.
Week 2 — equations and manufactured tests: derive the key relation and test inputs with known answers.
Week 3 — implementation and evidence: use a trusted library, record parameters and compare with a transparent baseline.
Week 4 — transfer and critique: apply the mechanism to a new context, report limits and identify when a simpler or different method is better.

Guidance for parents and educators

Ask ‘what stays true?’ and ‘what could make this conclusion fail?’ before asking for speed. Invite the student to draw the input and output, label units, estimate the answer and locate the first step that disagrees with a reference. These prompts reveal conceptual understanding more clearly than memorised terminology.

Advanced enrichment should sit beside secure school foundations in algebra, functions, geometry, statistics and computational thinking. It need not accelerate a student into a fixed career identity. A healthy goal is curiosity with discipline: the learner becomes comfortable reading unfamiliar notation, checking an example, revising a model and communicating uncertainty.

Evidence prompts for a useful conversation

Which quantity is observed and which is inferred?
Which step saves work, and why is it allowed?
Is the claim exact, approximate, probabilistic or empirical?
What units and conventions are in use?
Which input makes the method struggle?
What simple baseline could check the answer?
What would count as evidence that the chosen parameters are poor?

Mini-project assessment

A strong mini-project contains the original question, a small reproducible dataset, a hand-worked example, code or a spreadsheet trace, at least one boundary test, a comparison with a baseline, and a paragraph on limits. Assess the reasoning trail as well as the final output. If a student can explain why a surprising result appeared and repair the model, that is valuable mathematical progress even when the first attempt was wrong.

Ten evidence checks

Check 1: Calculate and plot regression residuals

Prepare a one-page evidence card for this capability. Put the definition or formula at the top, a hand-worked example in the middle, and a deliberately awkward input at the bottom. Beside each line, state what would make the result wrong: a convention, unit, index, random choice, approximation or modelling assumption. Reproduce the answer with an independent baseline and explain any difference before moving on. The card is complete only when another learner can follow it without guessing hidden parameters.

Check 2: Square residuals and retain observation alignment

Check 3: Fit an auxiliary regression

Check 4: Compute r-squared from sums of squares

Check 5: Calculate the nr-squared lm statistic

Check 6: Count degrees of freedom correctly

Check 7: Distinguish classic and koenker variants

Check 8: Separate mean and variance misspecification

Check 9: Choose between robust inference and variance modelling

Check 10: Report diagnostic uncertainty without causal overclaiming

Frequently asked questions

What does the Breusch–Pagan test examine?

It tests whether residual variance is systematically related to selected explanatory variables. State the relevant assumptions and parameters whenever the answer is used in a real calculation.

Why are residuals squared?

Squaring converts signed residuals into a positive proxy for error magnitude. State the relevant assumptions and parameters whenever the answer is used in a real calculation.

What is the auxiliary regression?

It regresses squared residuals on variables proposed to explain the variance. State the relevant assumptions and parameters whenever the answer is used in a real calculation.

Why is LM equal to nR-squared?

That form arises from the score-test derivation under the null and regularity conditions. State the relevant assumptions and parameters whenever the answer is used in a real calculation.

What are the degrees of freedom?

Usually the number of variance regressors excluding the constant. State the relevant assumptions and parameters whenever the answer is used in a real calculation.

What is the Koenker version?

It is a studentised variant that is less dependent on normal-error assumptions. State the relevant assumptions and parameters whenever the answer is used in a real calculation.

Does rejection mean OLS coefficients are biased?

Not automatically; exogeneity, efficiency and standard-error validity are separate issues. State the relevant assumptions and parameters whenever the answer is used in a real calculation.

Should I transform the response?

Only when the transformed estimand and model are substantively appropriate. State the relevant assumptions and parameters whenever the answer is used in a real calculation.

What should students know first?

Least squares, residuals, variance, R-squared, chi-square distributions and model assumptions. State the relevant assumptions and parameters whenever the answer is used in a real calculation.

What broader lesson does it teach?

A model assumption can be translated into an auxiliary measurable pattern, but diagnosis and remedy remain separate decisions. State the relevant assumptions and parameters whenever the answer is used in a real calculation.

Useful next reading and sources

Begin with Breusch and Pagan's 1979 Econometrica paper for the historical result and statsmodels' official het_breuschpagan documentation for a maintained modern interface or reference. Then connect the mechanism to these verified eduKateSG articles:

For the broader series, continue at eduKateSG Mathematics. Use primary sources for definitions and results, official documentation for present software behaviour, and experiments for performance on the actual task. Those evidence types support different claims and should not be merged casually.

Final takeaway

Why is mathematics important in this topic? Because it makes an invisible mechanism inspectable. Definitions say what the objects are. Equations show how information moves. Proof ideas explain what may safely be reused or approximated. Worked examples catch mistakes, while limits prevent a useful method from becoming an exaggerated promise. That combination of optimism and care is one of the lasting benefits of mathematics education.

Why Mathematics? | Breusch–Pagan Residual Variance, Auxiliary Regression and Heteroscedasticity