A crude 2×2 table can mix groups that differ in age, location, baseline risk or another variable. Stratification separates the data into comparable layers. Cochran–Mantel–Haenszel methods use the layer-specific counts to test conditional association and, for suitable 2×2 tables, estimate a common odds ratio. Mathematics matters because weighted sums, conditional expectations and sampling assumptions distinguish an adjusted summary from a misleading pooled comparison. The method does not make observational evidence causal, and a common summary is inappropriate when effects differ materially across strata. Those limits are part of the analysis, not fine print.
The practical search question is not simply “what is this named method?” but “which mathematical structure makes the method work, what does it guarantee, and what evidence would show that it is a poor choice?” This article keeps those questions together so students can connect mathematics education with computing, engineering, data reasoning and careful everyday decisions without turning one technique into a career promise.
Quick navigation: Why it matters · Mechanism · Worked example · Deeper ideas · Investigations · Limits · Learning route · Parent guidance · FAQs · Sources
Why this mathematics matters
Mathematics is important here because it turns an apparently visual or computational trick into a chain of reasons. The representation tells us what information is kept. The equation or invariant explains each legal step. The complexity analysis describes the work. The boundary conditions say when a conclusion should not be trusted. That combination is a practical form of problem-solving literacy.
A strong learner should be able to move in both directions: from a real task to a mathematical model, and from a formal guarantee back to its real meaning. Naming an algorithm is not enough. The useful achievement is to explain why the next operation is valid, check a small instance independently, and notice which assumptions the implementation has added.
Source: CDC Epi Info guide to stratified tables and Mantel–Haenszel estimates.
Primary or implementation reference: Mantel and Haenszel’s 1959 paper on retrospective-study analysis.
How the mechanism works
For each stratum i, arrange counts as a_i for exposed cases, b_i for exposed non-cases or controls, c_i for unexposed cases, and d_i for unexposed non-cases or controls; let n_i be the stratum total. The Mantel–Haenszel common odds-ratio estimate is theta_MH=[sum_i(a_i d_i/n_i)]/[sum_i(b_i c_i/n_i)]. The CMH association test compares observed a_i values with their expectations under conditional independence, weighting by their conditional variances. Before reporting a common effect, inspect stratum-specific odds ratios and test or reason about heterogeneity. Sparse counts, matched designs and risk-ratio targets require the version appropriate to the study design.
An odds ratio compares odds, not probabilities: OR=(a/b)/(c/d)=ad/bc for one 2×2 table. Stratification conditions on a third variable so the association between exposure and outcome is assessed within its levels. The Mantel–Haenszel weights arise from cross-products scaled by n_i, preventing a simple unweighted average of stratum odds ratios. Under the null and regularity conditions, the CMH statistic has an approximate chi-square distribution with one degree of freedom for the basic 2×2×K setting. Similarity of stratum effects supports a common odds-ratio summary; important heterogeneity or interaction means the separate effects may be the main story.
Keep a four-column notebook labelled object, rule, guarantee and caveat. In the first column name the inputs and stored state. In the second write one update precisely. In the third state what remains true after the update. In the fourth record a tie, approximation, domain restriction or hardware assumption. This routine prevents symbols from drifting away from their meaning.
Worked example and core calculations
Consider two deliberately simple strata. In stratum 1 let a=30, b=20, c=20, d=30 and n=100, giving OR=30×30/(20×20)=2.25. In stratum 2 let a=18, b=12, c=12, d=18 and n=60, also giving OR=2.25. The Mantel–Haenszel numerator is 30×30/100 + 18×18/60 = 9+5.4=14.4. The denominator is 20×20/100 + 12×12/60 = 4+2.4=6.4. Thus theta_MH=14.4/6.4=2.25. The crafted strata are homogeneous, so the common summary is transparent. This illustrative arithmetic is not evidence about any real exposure or health outcome.
How to audit the calculation
- Recompute every intermediate value directly from the definition.
- Keep labels, indices, signs, units and normalisation visible.
- Test a case with a known exact answer.
- State tie rules, random seeds, tolerances and boundary conventions.
- Compare with a slower transparent baseline or trusted implementation.
- Separate agreement of numbers from suitability of the real-world model.
Small examples are not childish; they are diagnostic instruments. A five-line trace can reveal a reversed inequality, a missing scale factor or an inconsistent label convention before a large dataset hides the same error behind plausible output.
Four deeper ideas
Stratification changes the comparison set
The question becomes whether exposure and outcome are associated among observations at the same stratum level.
To transfer this idea, ask what would remain true if labels, scale or input order changed. Then identify the step whose justification would fail if the stated assumption were removed. This turns the paragraph into a reusable reasoning pattern rather than a fact to memorise.
Weights come from table information
The common estimate is a ratio of weighted cross-products, not the arithmetic mean of odds ratios.
To transfer this idea, ask what would remain true if labels, scale or input order changed. Then identify the step whose justification would fail if the stated assumption were removed. This turns the paragraph into a reusable reasoning pattern rather than a fact to memorise.
Did you know? Adjustment can reverse a crude pattern
Unequal mixing across strata can create Simpson-type reversals, which is why the pooled table must be inspected critically.
To transfer this idea, ask what would remain true if labels, scale or input order changed. Then identify the step whose justification would fail if the stated assumption were removed. This turns the paragraph into a reusable reasoning pattern rather than a fact to memorise.
Heterogeneity can be the result
If stratum effects differ, forcing one number may hide interaction rather than remove confounding.
To transfer this idea, ask what would remain true if labels, scale or input order changed. Then identify the step whose justification would fail if the stated assumption were removed. This turns the paragraph into a reusable reasoning pattern rather than a fact to memorise.
Twelve practical investigations
1. Homogeneous strata
Use the worked tables. Calculate every cross-product.
What the mathematics reveals: A common estimate is interpretable here. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: The numbers are illustrative. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
2. Crude confounding
Mix strata in unequal proportions. Compare crude and adjusted estimates.
What the mathematics reveals: Composition can distort a pooled table. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: Explain the stratifying variable. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
3. Simpson reversal
Construct opposite pooled and within-stratum patterns. Trace weights.
What the mathematics reveals: Aggregation can reverse direction. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: Do not treat the paradox as magic. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
4. Effect modification
Use clearly different stratum odds ratios. Report them separately.
What the mathematics reveals: Interaction may be substantive. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: Do not hide it in one summary. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
5. Sparse cells
Create small expected counts. Review exact alternatives.
What the mathematics reveals: Asymptotic tests can be unreliable. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: Avoid automatic continuity fixes. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
6. Zero cell
Set one count to zero. Inspect estimator behaviour.
What the mathematics reveals: A finite summary may fail. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: Use justified methods. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
7. Odds versus risk
Convert probabilities to odds. Compare OR and risk ratio.
What the mathematics reveals: They answer different questions. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: Rare-outcome approximation is conditional. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
8. Broad strata
Combine a wide age range. Subdivide and compare.
What the mathematics reveals: Residual confounding can remain. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: Overstratification also loses precision. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
9. Misclassification
Move observations between cells. Recompute estimates.
What the mathematics reveals: Measurement quality affects association. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: Adjustment cannot recover unknown truth. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
10. Software audit
Run CDC or R output. Match table orientation.
What the mathematics reveals: Row order can invert the reported OR. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: Read labels, not only numbers. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
11. Confidence interval
Calculate with trusted software. Interpret width and scale.
What the mathematics reveals: Precision belongs beside the point estimate. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: Do not equate 95% with certainty. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
12. Study report
Write design, strata and estimand first. Separate facts and inference.
What the mathematics reveals: Statistical output needs context. Evidence to collect: write the input, predict the next state, carry out the calculation, and compare the result with an independently produced reference. Explain the exact invariant, equation, inequality or counting rule that supports the conclusion. Then vary one assumption and record whether the result changes smoothly, discontinuously or not at all.
Limit to keep visible: Avoid causal language without design support. A useful report places this boundary beside the result rather than hiding it in a final disclaimer.
Limits and misconceptions
CMH adjustment handles only measured, correctly classified stratification variables included in the analysis. Residual confounding can remain within broad strata. Sparse cells can weaken chi-square approximations, and zero counts require appropriate exact or model-based treatment rather than arbitrary deletion. Odds ratios are not risk ratios except under specific interpretations; the rare-outcome approximation has conditions. Selection bias, measurement error and reverse causation are not repaired by stratification. A common effect assumes sufficient homogeneity. Observational associations do not establish a causal mechanism, and medical decisions require qualified evidence beyond a classroom calculation.
Five misconceptions to challenge
- Adjustment proves causation: It addresses a specified confounding structure, not every bias or causal question.
- The common odds ratio is a simple average: It is a ratio of weighted cross-products.
- Odds and probability are the same: Odds are p/(1−p).
- Different stratum effects should always be pooled: Heterogeneity may make separate estimates more informative.
- A non-significant test proves no association: It may reflect limited precision and does not prove exact equality.
Another broad misconception is that advanced mathematics automatically creates intelligence, admission, income or employment. It does not. Deliberate study can strengthen modelling, abstraction, calculation and explanation, while real pathways also depend on interests, communication, domain knowledge, opportunities and circumstances outside one topic.
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 software returned a number.
Step 1: Construct stratified 2×2 tables
Begin with the smallest non-trivial instance that lets you construct stratified 2×2 tables. Predict the result before calculating. Write each state change clearly enough that another student can reproduce it. Compare with a trusted implementation only after the hand trace is complete.
Next, change one feature—size, ordering, sign, tolerance, labelling or data distribution—and explain which part of the reasoning changes. Finish by teaching the step without code and by naming one input for which the method is unhelpful. This predict–calculate–vary–explain cycle is a strong test of transferable understanding.
Step 2: Calculate stratum odds ratios
Begin with the smallest non-trivial instance that lets you calculate stratum odds ratios. Predict the result before calculating. Write each state change clearly enough that another student can reproduce it. Compare with a trusted implementation only after the hand trace is complete.
Next, change one feature—size, ordering, sign, tolerance, labelling or data distribution—and explain which part of the reasoning changes. Finish by teaching the step without code and by naming one input for which the method is unhelpful. This predict–calculate–vary–explain cycle is a strong test of transferable understanding.
Step 3: Calculate a mantel–haenszel estimate
Begin with the smallest non-trivial instance that lets you calculate a Mantel–Haenszel estimate. Predict the result before calculating. Write each state change clearly enough that another student can reproduce it. Compare with a trusted implementation only after the hand trace is complete.
Next, change one feature—size, ordering, sign, tolerance, labelling or data distribution—and explain which part of the reasoning changes. Finish by teaching the step without code and by naming one input for which the method is unhelpful. This predict–calculate–vary–explain cycle is a strong test of transferable understanding.
Step 4: Distinguish odds and risk
Begin with the smallest non-trivial instance that lets you distinguish odds and risk. Predict the result before calculating. Write each state change clearly enough that another student can reproduce it. Compare with a trusted implementation only after the hand trace is complete.
Next, change one feature—size, ordering, sign, tolerance, labelling or data distribution—and explain which part of the reasoning changes. Finish by teaching the step without code and by naming one input for which the method is unhelpful. This predict–calculate–vary–explain cycle is a strong test of transferable understanding.
Step 5: Compare crude and adjusted estimates
Begin with the smallest non-trivial instance that lets you compare crude and adjusted estimates. Predict the result before calculating. Write each state change clearly enough that another student can reproduce it. Compare with a trusted implementation only after the hand trace is complete.
Next, change one feature—size, ordering, sign, tolerance, labelling or data distribution—and explain which part of the reasoning changes. Finish by teaching the step without code and by naming one input for which the method is unhelpful. This predict–calculate–vary–explain cycle is a strong test of transferable understanding.
Step 6: Inspect effect heterogeneity
Begin with the smallest non-trivial instance that lets you inspect effect heterogeneity. Predict the result before calculating. Write each state change clearly enough that another student can reproduce it. Compare with a trusted implementation only after the hand trace is complete.
Next, change one feature—size, ordering, sign, tolerance, labelling or data distribution—and explain which part of the reasoning changes. Finish by teaching the step without code and by naming one input for which the method is unhelpful. This predict–calculate–vary–explain cycle is a strong test of transferable understanding.
Step 7: Interpret a conditional association test
Begin with the smallest non-trivial instance that lets you interpret a conditional association test. Predict the result before calculating. Write each state change clearly enough that another student can reproduce it. Compare with a trusted implementation only after the hand trace is complete.
Next, change one feature—size, ordering, sign, tolerance, labelling or data distribution—and explain which part of the reasoning changes. Finish by teaching the step without code and by naming one input for which the method is unhelpful. This predict–calculate–vary–explain cycle is a strong test of transferable understanding.
Step 8: Recognise sparse-table limits
Begin with the smallest non-trivial instance that lets you recognise sparse-table limits. Predict the result before calculating. Write each state change clearly enough that another student can reproduce it. Compare with a trusted implementation only after the hand trace is complete.
Next, change one feature—size, ordering, sign, tolerance, labelling or data distribution—and explain which part of the reasoning changes. Finish by teaching the step without code and by naming one input for which the method is unhelpful. This predict–calculate–vary–explain cycle is a strong test of transferable understanding.
Step 9: Audit software table orientation
Begin with the smallest non-trivial instance that lets you audit software table orientation. Predict the result before calculating. Write each state change clearly enough that another student can reproduce it. Compare with a trusted implementation only after the hand trace is complete.
Next, change one feature—size, ordering, sign, tolerance, labelling or data distribution—and explain which part of the reasoning changes. Finish by teaching the step without code and by naming one input for which the method is unhelpful. This predict–calculate–vary–explain cycle is a strong test of transferable understanding.
Step 10: Separate association from causation
Begin with the smallest non-trivial instance that lets you separate association from causation. Predict the result before calculating. Write each state change clearly enough that another student can reproduce it. Compare with a trusted implementation only after the hand trace is complete.
Next, change one feature—size, ordering, sign, tolerance, labelling or data distribution—and explain which part of the reasoning changes. Finish by teaching the step without code and by naming one input for which the method is unhelpful. This predict–calculate–vary–explain cycle is a strong test of transferable understanding.
A four-week practice plan
- Week 1 — vocabulary and diagrams: define each object, redraw the 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 method to a new context, report limits and identify when a 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, estimate the result 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 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. Repairing a model after a surprising result is valuable mathematical progress.
Six evidence checks for a finished project
Check 1: Construct stratified 2×2 tables
Ask the student to demonstrate how to construct stratified 2×2 tables without relying on a hidden software step. The evidence should include a labelled input, the exact rule used, one intermediate state and a result that can be checked independently. If the answer changes after labels or order change, the student should say whether that reflects the mathematics or only a convention.
Then request one deliberately awkward example. The learner should predict the difficulty before running the method, explain what the output does and does not establish, and name a reasonable alternative or safeguard. This check joins correctness with judgement: a technically valid calculation is useful only when its assumptions and interpretation remain visible.
Check 2: Calculate stratum odds ratios
Ask the student to demonstrate how to calculate stratum odds ratios without relying on a hidden software step. The evidence should include a labelled input, the exact rule used, one intermediate state and a result that can be checked independently. If the answer changes after labels or order change, the student should say whether that reflects the mathematics or only a convention.
Then request one deliberately awkward example. The learner should predict the difficulty before running the method, explain what the output does and does not establish, and name a reasonable alternative or safeguard. This check joins correctness with judgement: a technically valid calculation is useful only when its assumptions and interpretation remain visible.
Check 3: Calculate a mantel–haenszel estimate
Ask the student to demonstrate how to calculate a Mantel–Haenszel estimate without relying on a hidden software step. The evidence should include a labelled input, the exact rule used, one intermediate state and a result that can be checked independently. If the answer changes after labels or order change, the student should say whether that reflects the mathematics or only a convention.
Then request one deliberately awkward example. The learner should predict the difficulty before running the method, explain what the output does and does not establish, and name a reasonable alternative or safeguard. This check joins correctness with judgement: a technically valid calculation is useful only when its assumptions and interpretation remain visible.
Check 4: Distinguish odds and risk
Ask the student to demonstrate how to distinguish odds and risk without relying on a hidden software step. The evidence should include a labelled input, the exact rule used, one intermediate state and a result that can be checked independently. If the answer changes after labels or order change, the student should say whether that reflects the mathematics or only a convention.
Then request one deliberately awkward example. The learner should predict the difficulty before running the method, explain what the output does and does not establish, and name a reasonable alternative or safeguard. This check joins correctness with judgement: a technically valid calculation is useful only when its assumptions and interpretation remain visible.
Check 5: Compare crude and adjusted estimates
Ask the student to demonstrate how to compare crude and adjusted estimates without relying on a hidden software step. The evidence should include a labelled input, the exact rule used, one intermediate state and a result that can be checked independently. If the answer changes after labels or order change, the student should say whether that reflects the mathematics or only a convention.
Then request one deliberately awkward example. The learner should predict the difficulty before running the method, explain what the output does and does not establish, and name a reasonable alternative or safeguard. This check joins correctness with judgement: a technically valid calculation is useful only when its assumptions and interpretation remain visible.
Check 6: Inspect effect heterogeneity
Ask the student to demonstrate how to inspect effect heterogeneity without relying on a hidden software step. The evidence should include a labelled input, the exact rule used, one intermediate state and a result that can be checked independently. If the answer changes after labels or order change, the student should say whether that reflects the mathematics or only a convention.
Then request one deliberately awkward example. The learner should predict the difficulty before running the method, explain what the output does and does not establish, and name a reasonable alternative or safeguard. This check joins correctness with judgement: a technically valid calculation is useful only when its assumptions and interpretation remain visible.
Frequently asked questions
What does CMH stand for?
Cochran–Mantel–Haenszel.
Why stratify data?
To compare exposure and outcome within levels of a potential confounder or design variable.
What is the common odds ratio?
A weighted summary of stratum-specific 2×2 associations when a common effect is defensible.
Is it an arithmetic average?
No.
What does the CMH test assess?
Conditional association under its stated sampling assumptions.
What if stratum odds ratios differ greatly?
Report and investigate heterogeneity or interaction instead of relying on one common effect.
Is an odds ratio a risk ratio?
No.
Does adjustment prove causation?
No.
What if cells are sparse?
Use appropriate exact, model-based or carefully justified alternatives.
What should students know first?
Fractions, probability, odds, contingency tables, weighted sums and study design.
Useful next reading and sources
Source: CDC Epi Info guide to stratified tables and Mantel–Haenszel estimates.
Source: Mantel and Haenszel’s 1959 paper on retrospective-study analysis.
Related eduKateSG reading: CUSUM change detection.
Related eduKateSG reading: PAVA isotonic regression.
Related eduKateSG reading: Comparing percentages fairly.
Read sources with a question in mind: which definition, theorem, default or limitation is being supported? An authoritative link is useful evidence, but it does not replace a worked calculation or a clear statement of how the source applies to the example.
A cheerful final thought
Mathematics becomes friendlier when every symbol has a job and every claim has a boundary. Start small, make the state visible, test the awkward case, and celebrate the moment when a pattern becomes explainable. That is the practical benefit of learning mathematics: not certainty about every future, but a growing ability to ask better questions and build trustworthy answers.