Cochran–Mantel–Haenszel statistics explained

In statistics, the Cochran–Mantel–Haenszel test (CMH) is a test used in the analysis of stratified or matched categorical data. It allows an investigator to test the association between a binary predictor or treatment and a binary outcome such as case or control status while taking into account the stratification.[1] Unlike the McNemar test, which can only handle pairs, the CMH test handles arbitrary strata sizes. It is named after William G. Cochran, Nathan Mantel and William Haenszel.[2] [3] Extensions of this test to a categorical response and/or to several groups are commonly called Cochran–Mantel–Haenszel statistics.[4] It is often used in observational studies in which random assignment of subjects to different treatments cannot be controlled but confounding covariates can be measured.

Definition

We consider a binary outcome variable such as case status (e.g. lung cancer) and a binary predictor such as treatment status (e.g. smoking). The observations are grouped in strata. The stratified data are summarized in a series of 2 × 2 contingency tables, one for each stratum. The i-th such contingency table is:

Treatment No treatment Row total
Case Ai Bi N1i
Controls Ci Di N2i
Column total M1i M2i Ti

The common odds-ratio of the K contingency tables is defined as:

R=

K
{{\sum
i=1
AiDi
Ti
} \over },The null hypothesis is that there is no association between the treatment and the outcome. More precisely, the null hypothesis is

H0:R=1

and the alternative hypothesis is

H1:R\ne1

. The test statistic is:

\xiCMH=

\left[
K
\sum
i=1
\left(Ai-
N1iM1i
Ti
\right)\right]2
K
\sum{N1iN2iM1iM2i\over
2(T
T
i-1)
i=1
}.It follows a

\chi2

distribution asymptotically with 1 df under the null hypothesis.

Subset stability

The standard odds- or risk ratio of all strata could be calculated, giving risk ratios

r1,r2,...,rn

, where

n

is the number of strata. If the stratification were removed, there would be one aggregate risk ratio of the collapsed table; let this be

R

.

One generally expects the risk of an event unconditional on the stratification to be bounded between the highest and lowest risk within the strata (or identically with odds ratios).It is easy to construct examples where this is not the case, and

R

is larger or smaller than all of

ri

for

i\in1,...,n

.This is comparable but not identical to Simpson's paradox, and as with Simpson's paradox, it is difficult to interpret the statistic and decide policy based upon it.

Klemens[5] defines a statistic to be subset stable iff

R

is bounded between

min(ri)

and

max(ri)

, and a well-behaved statistic as being infinitely differentiable and not dependent on the order of the strata.Then the CMH statistic is the unique well-behaved statistic satisfying subset stability.

Related tests

External links

Notes and References

  1. Book: Agresti, Alan . 2002 . Categorical Data Analysis . Hoboken, New Jersey . John Wiley & Sons, Inc. . 231–232 . 0-471-36093-7.
  2. William G. Cochran. Some Methods for Strengthening the Common χ2 Tests . Biometrics . December 1954 . 10 . 4 . 417–451 . 3001616 . 10.2307/3001616.
  3. Nathan Mantel and William Haenszel . Statistical aspects of the analysis of data from retrospective studies of disease . Journal of the National Cancer Institute . April 1959 . 22 . 4 . 719–748 . 13655060 . 10.1093/jnci/22.4.719 .
  4. Nathan Mantel . Chi-Square Tests with One Degree of Freedom, Extensions of the Mantel–Haenszel Procedure . Journal of the American Statistical Association . September 1963 . 58 . 303 . 690–700 . 2282717 . 10.1080/01621459.1963.10500879.
  5. An Analysis of U.S. Domestic Migration via Subset-stable Measures of Administrative Data . Ben Klemens. Journal of Computational Social Science . June 2021 . 5. 351–382. 10.1007/s42001-021-00124-w. 236308711. subscription.
  6. Book: Agresti, Alan . 2002 . Categorical Data Analysis . Hoboken, New Jersey . John Wiley & Sons, Inc. . 413 . 0-471-36093-7.
  7. Testing hypotheses in case-control studies-equivalence of Mantel–Haenszel statistics and logit score tests.. Day N.E., Byar D.P.. Biometrics . 35 . 3 . 623–630 . September 1979 . 2530253. 10.2307/2530253. 497345.