Recommended tests for association in 2×2 tables

Unit for Applied Clinical Research, Department of Cancer Research and Molecular Medicine, Norwegian University of Science and Technology, Trondheim, Norway.
Statistics in Medicine (Impact Factor: 1.83). 03/2009; 28(7):1159-75. DOI: 10.1002/sim.3531
Source: PubMed


The asymptotic Pearson's chi-squared test and Fisher's exact test have long been the most used for testing association in 2x2 tables. Unconditional tests preserve the significance level and generally are more powerful than Fisher's exact test for moderate to small samples, but previously were disadvantaged by being computationally demanding. This disadvantage is now moot, as software to facilitate unconditional tests has been available for years. Moreover, Fisher's exact test with mid-p adjustment gives about the same results as an unconditional test. Consequently, several better tests are available, and the choice of a test should depend only on its merits for the application involved. Unconditional tests and the mid-p approach ought to be used more than they now are. The traditional Fisher's exact test should practically never be used.

Download full-text


Available from: Morten Wang Fagerland,
    • "Clearly, there are other (asymptotic or non-asymptotic) frequentist test approaches which are under certain assumptions on the expected cell counts more robust than chi-square tests; see, e.g. Lydersen et al. (2009) for a biostatistics tutorial with practical guidelines for choosing a marginal testing strategy in the case of a (2 × 2)-table. However, an automated application of such guidelines for a large number of contingency tables simultaneously, where parameters like the expected minor allele frequency are prone to change considerably from one genomic position to the other, appears extremely challenging. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Genetic association studies lead to simultaneous categorical data analysis. The sample for every genetic locus consists of a contingency table containing the numbers of observed genotype-phenotype combinations. Under case-control design, the row counts of every table are identical and fixed, while column counts are random. The aim of the statistical analysis is to test independence of the phenotype and the genotype at every locus. We present an objective Bayesian methodology for these association tests, which relies on the conjugacy of Dirichlet and multinomial distributions. Being based on the likelihood principle, the Bayesian tests avoid looping over all tables with given marginals. Making use of data generated by The Wellcome Trust Case Control Consortium (WTCCC), we illustrate that the ordering of the Bayes factors shows a good agreement with that of frequentist p-values. Furthermore, we deal with specifying prior probabilities for the validity of the null hypotheses, by taking linkage disequilibrium structure into account and exploiting the concept of effective numbers of tests. Application of a Bayesian decision theoretic multiple test procedure to the WTCCC data illustrates the proposed methodology. Finally, we discuss two methods for reconciling frequentist and Bayesian approaches to the multiple association test problem.
    Statistical Applications in Genetics and Molecular Biology 07/2015; 14(4). DOI:10.1515/sagmb-2014-0052 · 1.13 Impact Factor
  • Source
    • "Continuous variables were compared between groups with the two-sample t-test (if approximately normally distributed) or the Mann–Whitney U-test (if markedly skewed). Categorical variables were compared using the Fisher mid-P-test [33]. Spearman rank correlation was used to assess simple (univariate) associations of continuous variables with MRP-8/14 or CRP. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Background Obstructive sleep apnea (OSA) and obesity are often present concomitantly. Their potential contribution to inflammation remains an ongoing debate. The objectives of this study were to investigate whether variables of sleep-disordered breathing are associated with levels of myeloid-related protein-8/14 (MRP-8/14) or C-reactive protein (CRP) and to characterize how adiposity interacts with these associations in individuals evaluated for possible OSA. Methods Consecutive individuals referred to Lovisenberg Diakonale Hospital’s sleep laboratory between 1st October 2009 and 1st March 2010 were included. We characterized the biomarker distribution sampled the morning after sleep and related these to clinical characteristics and variables recorded during polygraphy or polysomnography. Results Of the total study population of 222 individuals, 161 (72.5%) were diagnosed with OSA (apnea–hypopnea index (AHI) index ⩾5/h). In baseline models (multiple median regression adjusted for age and sex), AHI was independently associated with MRP-8/14 (P = 0.025) and CRP (P < 0.001). The associations were attenuated after the addition of body mass index (BMI), but remained statistically significant for CRP (P = 0.025). However, in final models adjusted for additional factors (systolic blood pressure, cholesterol:high-density lipoprotein ratio, glycosylated haemoglobin, smoking, and cardiovascular disease), only average oxygen saturation for MRP-8/14 (P = 0.028) and oxygen desaturation index (ODI) for CRP (P = 0.037) remained independent predictors of inflammation, whereas AHI lost its predictive value (MRP-8/14; P = 0.30 and CRP; P = 0.092). The association between several variables of sleep-disordered breathing and inflammation were stronger in individuals with a higher BMI (P for interaction < 0.05 for AHI, nadir oxygen saturation, and time <90% oxygen saturation). Conclusions No definitive indication of independent immunological activity resulting from apneas and hypopneas was found in final models adjusted for other factors associated with inflammation, whereas average oxygen saturation for MRP-8/14 and ODI for CRP remained statistically significant predictors. Interactions were observed between BMI and several variables of sleep-disordered breathing on MRP-8/14 and CRP levels.
    Sleep Medicine 07/2014; 15(7). DOI:10.1016/j.sleep.2014.03.008 · 3.15 Impact Factor
  • Source
    • "However , for testing the equality of two proportions, or the risk difference, there are multiple formulae, which can yield very different results. More than 60 asymptotic tests and many exact tests have been considered for 2 Â 2 tables [4]. Many researchers may not be aware that there are multiple ways to calculate sample size for 2 Â 2 table designs, and even with this awareness, they may not know which approach and assumptions their software uses. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Several methods exist to calculate sample size for the difference of proportions (risk difference). Researchers are often unaware that there are different formulae, different underlying assumptions, and what the impact of choice of formula is on the calculated sample size. The aim of this study was to discuss and compare different sample size formulae for the risk difference. Four sample size formulae were used to calculate sample size for nine scenarios. Software documentation for SAS, Stata, G*Power, PASS, StatXact, and several R libraries were searched for default assumptions. Each package was used to calculate sample size for two scenarios. We demonstrate that for a set of parameters, sample size can vary as much as 60% depending on the formula used. Varying software and assumptions yielded discrepancies of 78% and 7% between the smallest and largest calculated sizes, respectively. Discrepancies were most pronounced when powering for large risk differences. The default assumptions varied considerably between software packages, and defaults were not clearly documented. Researchers should be aware of the assumptions in power calculations made by different statistical software packages. Assumptions should be explicitly stated in grant proposals and manuscripts and should match proposed analyses.
    Journal of clinical epidemiology 01/2014; 67(5). DOI:10.1016/j.jclinepi.2013.10.008 · 3.42 Impact Factor
Show more