Pearson's Chi-Squared Test¶
A categorical-count test that compares observed and null-expected frequencies through Pearson's weighted squared residuals.
Core Idea¶
The Pearson chi-squared test checks a categorical-count null by comparing each observed cell count with the count expected if that null held. It sums squared differences divided by their positive expected counts, then evaluates the total under a chi-square reference appropriate to the design. A small tail probability challenges the null; a large one leaves it unrefuted by this sample and procedure. The name identifies the count statistic and its calibrated use, not every test whose reference happens to be chi-square.[ref-9e1a92be9249][ref-d705e0cc198e]
Scope of Application¶
The test applies when a categorical null supplies defensible expected cell counts and a valid reference comparison.
- Binned distributional fit: NIST checks 101 aluminum-alloy fatigue lives against a fitted three-parameter Weibull law, reporting p = 0.15.[^ref-9e1a92be9249]
- Contingency independence: NIST's constructed shift-by-defect wafer table yields statistic 19.18 on six reference degrees of freedom and p = 0.00387.[^ref-d705e0cc198e]
Clarity¶
Report what one count represents, the tested null, included cells, how expectations were obtained, the statistic and its reference degrees of freedom. The sampling unit, mutually exclusive bins and parameter-estimation procedure determine which counts and reference apply. For independence, an expected cell count comes from its row total times column total divided by the grand total; a fitted distribution supplies bin probabilities instead.[ref-9e1a92be9249][ref-d705e0cc198e] These are distinct nulls sharing Pearson's arithmetic. Non-rejection does not prove the model correct, and rejection does not identify a cause or every influential cell.
Manages Complexity¶
The weighted sum compresses many cell discrepancies into one omnibus test, making a whole-table question tractable. That compression hides the signs and locations of deviations. Examine cell residuals when a table-level result needs explanation; descriptive inspection is different from running new cell-specific significance tests, which need separate calibration. Bin choices and small expected counts can change what the summary detects and weaken its chi-square approximation; regrouping or an appropriate alternative requires design-specific judgment.[ref-9e1a92be9249][ref-55bad3dbd9dc]
Abstract Reasoning¶
Start with a declared categorical null and derive positive expected counts for its cells. Compute the sum of each squared observed-minus-expected difference divided by that cell's expectation; then use the reference derived for that sampling and estimation design. In NIST's wafer example, three shifts and four defect types give six independence degrees of freedom. Its Weibull example instead adjusts the binned-fit reference for parameters estimated from bin counts. The same statistic does not make those calibrations interchangeable.[ref-9e1a92be9249][ref-d705e0cc198e]
Knowledge Transfer¶
The relation transfers literally between distributional fit and contingency independence: null, expected counts, Pearson residual sum, reference and bounded conclusion. Goodness-of-Fit Test is the direct genus: model, discrepancy, calibration and bounded adequacy readout. Statistical Test is inherited through it. Comparison is internal to each observed-versus-expected cell term; that operation alone supplies neither categorical null nor p-value. At a new table, re-derive expectations, constraints and reference rather than copying the first setting's p-value or degrees of freedom.[ref-9e1a92be9249][ref-d705e0cc198e]
Relationships to Other Abstractions¶
Current abstraction Pearson's Chi-Squared Test Domain-specific
Parents (2) — more general patterns this builds on
-
Pearson's Chi-Squared Test is a kind of Goodness-of-Fit Test Domain-specific
Pearson's categorical-count procedure is a goodness-of-fit test with a fixed weighted residual statistic.
-
Pearson's Chi-Squared Test is part of Comparison Prime
Each Pearson residual term internally compares observed and null-expected counts for one cell.
Hierarchy paths (2) — routes to 2 parentless roots
- Pearson's Chi-Squared Test → Goodness-of-Fit Test → Statistical Test
- Pearson's Chi-Squared Test → Comparison
Neighborhood in Abstraction Space¶
Pearson's Chi-Squared Test sits in a sparse region of the domain-specific corpus (78th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Unclustered & Miscellaneous (2721 abstractions)
Nearest neighbors
- Log-Linear Analysis — 0.86
- Join Count Statistic — 0.84
- Phi Coefficient — 0.83
- Euler Characteristic — 0.82
- Counternull — 0.82
Computed from structural-signature embeddings · 2026-10-17
References¶
[^ref-9e1a92be9249]: NIST/SEMATECH, “Chi-Square Goodness-of-Fit Test,” e-Handbook of Statistical Methods, §1.3.5.15, especially Test Statistic, Critical Region and Application Example. https://www.itl.nist.gov/div898/handbook/eda/section3/eda35f.htm. Full official HTML consulted; the original fatigue-life experiment was not independently inspected.
[^ref-d705e0cc198e]: NIST/SEMATECH, “How can we compare the results of classifying according to several categories?,” e-Handbook of Statistical Methods, §7.4.5, especially Industrial example, Expected cell frequencies, Test statistic and Degrees of freedom. https://www.itl.nist.gov/div898/handbook/prc/section4/prc45.htm. Full official HTML consulted; the wafer data are presented as a constructed industrial example.
[^ref-55bad3dbd9dc]: NIST Dataplot, “Chi-Square Independence Test,” reference page, especially the Note paragraphs on expected frequencies and sampling designs. https://www.itl.nist.gov/div898/software/dataplot/refman1/auxillar/chistest.htm. Full official HTML consulted; its competing expected-cell guidance and exact-test options are design dependent.