Some Methods for Strengthening the Common χ 2 Tests
The DOI matches its Crossref record; 18 of the 18 references it cites were found in a registry or an index.
Abstract
Since the x2 tests of goodness of fit and of association in contingency tables are presented in many courses on statistical methods for beginners in the subject, it is not surprising that x2 has become one of the most commonly-used techniques, even by scientists who profess only a smattering of knowledge of statistics. It is also not surprising that the technique is sometimes misused, e.g. by calculating x2 from data that are not frequencies or by errors in counting the number of degrees of freedom. A good catalogue of mistakes of this kind has been given by Lewis and Burke (1). In this paper I want to discuss two kinds of failure to make the best use of x2 tests which I have observed from time to time in reading reports of biological research. The first arises because x2 tests, as has often been pointed out, are not directed against any specific alternative to the null hypothesis. In the computation of x2, the deviations (fi mi) between observed and expected frequencies are squared, divided by mi in order to equalize the variances (approximately), and added. No attempt is made to detect any particular pattern of deviations (fi mi) that may hold if the null hypothesis is false. One consequence is that the usual x2 tests are often insensitive, and do not indicate significant results when the null hypothesis is actually false. Some forethought about the kind of alternative hypothesis that is likely to hold may lead to alternative tests that are more powerful and appropriate. Further, when the ordinary x2 test does give a significant result, it does not direct attention to the way in which the null hypothesis disagrees with the data, although the pattern of deviations may be informative and suggestive for future research. The remedy here is to supplement the ordinary test by additional tests that help to reveal the significant type of deviation. In this paper a number of methods for strengthening or supplementing the most common uses of the ordinary x2 test will be presented and illustrated by numerical examples. The principal devices are as follows:
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“Cochran's [4] criterion for choice of test in r × c tables is extensively cited (more than 1800 times by 2006).It maintains that Pearson's asymptotic chi-square test is acceptable when all m i j >1 and no more than 20 per cent of the cells have m i j <5.Otherwise, Fisher's exact test is recommended.”
From Choice of test for association in small sample unordered r × c tables (2007) · cites this paper as [4]
Citations
Cites
The $\chi^2$ Test of Goodness of Fit
Cochran, W. G. (1952) · The Annals of Mathematical Statistics · doi:10.1214/aoms/1177729380 · Cited by 1,106
Self-citationFitting the Negative Binomial Distribution to Biological Data
Bliss, C. I., & Fisher, R. A. (1953) · Biometrics · doi:10.2307/3001850 · Cited by 1,020
Note on an Exact Treatment of Contingency, Goodness of Fit and Other Problems of Significance
Freeman, G. H., & Halton, J. H. (1951) · Biometrika · doi:10.2307/2332323 · Cited by 954
Some Difficulties of Interpretation Encountered in the Application of the Chi-Square Test
Berkson, J. (1938) · Journal of the American Statistical Association · doi:10.1080/01621459.1938.10502329 · Cited by 411
Sampling Theory of the Negative Binomial and Logarithmic Series Distributions
Anscombe, F. J. (1950) · Biometrika · doi:10.2307/2332388 · Cited by 404
Some Difficulties of Interpretation Encountered in the Application of the Chi-Square Test
Berkson, J. (1938) · Journal of the American Statistical Association · doi:10.2307/2279690 · Cited by 259
THE ANALYSIS OF CONTINGENCY TABLES WITH GROUPINGS BASED ON QUANTITATIVE CHARACTERS
YATES, F. (1948) · Biometrika · doi:10.1093/biomet/35.1-2.176 · Cited by 203
Analysis of Factorial Arrangements when the Data are Proportions
Dyke, G. V., & Patterson, H. D. (1952) · Biometrics · doi:10.2307/3001521 · Cited by 143
Short-Cut Formulas for the Exact Partition of χ 2 in Contingency Tables
Kimball, A. W. (1954) · Biometrics · doi:10.2307/3001617 · Cited by 125
The derivation and partition of chi2 in certain discrete distributions.
Lancaster, H. O. (1949) · PubMed · Cited by 118
Cited by
Items 1,001–1,013 of 1,013
Xue, Y. (2024) · Education and Information Technologies · doi:10.1007/s10639-024-12596-0 · Cited by 33
Santhosh, M., Farooqi, H., Ammar, M., Siby, N., Bhadra, J., Al-Thani, N. J., Sellami, A., Fatima, N., & Ahmad, Z. (2023) · Journal of Science Education and Technology · doi:10.1007/s10956-023-10063-y · Cited by 28
Menescardi, C., Falco, C., Ros, C., Morales-Sánchez, V., & Hernández-Mendo, A. (2019) · Frontiers in Psychology · doi:10.3389/fpsyg.2019.02708 · Cited by 23
Design characteristics that influence attrition in geriatric antidepressant trials: meta‐analysis
Heo, M., Papademetriou, E., & Meyers, B. S. (2009) · International Journal of Geriatric Psychiatry · doi:10.1002/gps.2211 · Cited by 18
“Nevertheless, to test significance of difference in attrition rates between placebo and antidepressants among placebo-controlled studies, we applied Cochran- Copyright # 2009 John Wiley & Sons, Ltd. Mantel- Haenszel (CMH; Cochran, 1954; Mantel and Haenszel, 1959) test.”
MethodOn the validity condition of the chi-squared test in 2×2 tables
Andrés, A. M., Sánchez Quevedo, M. J., Tapia García, J. M., & Silva-Mato, A. (2005) · Test · doi:10.1007/bf02595399 · Cited by 18
Polo-Ferrero, L., Navarro-López, V., Fuentes, M., Lacal, J., Cancelas-Felgueras, M. D., Santos-Blázquez, N., Méndez-Sánchez, R., & Sánchez-González, J. L. (2025) · Nursing Reports · doi:10.3390/nursrep15030089 · Cited by 13
Lachin, J. M., & Bautista, O. M. (1995) · Cancer Treatment and Research · doi:10.1007/978-1-4615-2009-2_10 · Cited by 12
Hyperpersonal feedback and online romance fraud: an empirical examination
Wang, F., & Dickinson, T. (2025) · Journal of Crime and Justice · doi:10.1080/0735648x.2024.2376700 · Cited by 9
“We chose Fisher’s Exact test because the Chi-Square test should not be used when the smallest expected number of the sample size is less than 5 and the total sample size is less than 20 (Cochran 1954; Kim 2017).”
Optimal testing strategies for large, sparse multinomial models
Maiste, P. J., & Weir, B. S. (2004) · Computational Statistics & Data Analysis · doi:10.1016/j.csda.2003.08.002 · Cited by 8
Chihuri, S., Blanchard, A., DiGuiseppi, C. G., & Li, G. (2024) · Journal of Autism and Developmental Disorders · doi:10.1007/s10803-023-06091-7 · Cited by 5
Hsing (1996) · doi:10.1002/(sici)1097-0215(19960208)65:4 · Cited by 1
Martín-Díaz, P., Carratalá-Tejada, M., Navarro-López, V., Fernández-González, P., & Cuesta-Gómez, A. (2026) · European Journal of Pediatrics · doi:10.1007/s00431-026-06871-0
Lippens, L., Vermeiren, S., & Baert, S. (2023) · European Economic Review · doi:10.1016/j.euroecorev.2022.104315
Metadata
- Authors
- Cochran, W. G.
- Corresponding author
- Cochran, William G.
- Venue
- Biometrics
- Publisher
- JSTOR
- Published
- 1954-12
- Volume
- 10
- Issue
- 4
- Pages
- 417
- ISSN
- 0006-341X (ISSN-L)
- OpenAlex
- W2330813625
- Semantic Scholar
- e798116fb13f44ca8cb299b83da78a6d216042ae
- Cited by
- 3,718
- References
- 18
- Citations by year
- 2026: 118 · 2025: 117 · 2024: 136 · 2023: 119 · 2022: 111 · 2021: 120 · 2020: 88 · 2019: 97 · 2018: 100 · 2017: 89 · 2016: 99 · 2015: 69 · 2014: 105 · 2013: 95 · 2012: 80
- FWCI
- 17.49
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- Biometrics (publishedVersion)
- Topics
- Physical Sciences › Mathematics › Statistics and Probability › Probability and Statistical Research
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Methods and Applications
- Physical Sciences › Mathematics › Statistics and Probability › Statistical Distribution Estimation and Applications
- Keywords
- chi-square test, goodness of fit, contingency tables, statistical power, alternative hypothesis, misuse of statistical tests
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.