TY - JOUR
T1 - An evaluation of simple methods for the estimation of a common odds ratio in clusters with variable size
AU - Ahn, Chul
N1 - Funding Information:
I would like to thank Drs. J.N.K. Rao, Allan Donner, and Walter Hauck for their valuable suggestions. I would also like to thank two referees for a number of constructive suggestions. This research was partially supported by grants CA64775 and M01RR02558 from the National Institute of Health.
PY - 1997/3/6
Y1 - 1997/3/6
N2 - In this paper, the results from a simulation study for the estimation of a common odds ratio in multiple 2 × 2 tables when the data are correlated within clusters are presented. The size of clusters in each stratum is modelled by the negative binomial distribution truncated below 1 with mean cluster size, μ = 4, 5, and 6, and imbalance parameter, κ = 0.6, 0.8, and 1.0. The correlation of the data is modelled by the beta-binomial distribution. A simulation study is conducted to compare the performances of the Mantel-Haenszel estimator (ψ̂MH), the modified Donner-Hauck estimator (ψ̂MDH) and the Rao-Scott estimator (ψNRS, ψPRS) in terms of their biases, absolute biases, mean squared errors, and 95% coverage proportions. The modified Donner-Hauck estimator ψ̂MDH performs better than the other estimators in terms of the bias, absolute bias, and MSE when ψ ≥ 3 in the unbalanced design. In general, there are negligible differences in the bias, absolute bias, and MSE among the estimators when ψ = 1 or κ = 1.0. The estimator ψ̂MDH has generally coverage proportions closer to the nominal level than the other estimators for ρ ≥ 0.3 in the unbalanced design. When κ = 0.8, ψ̂PRS and ψ̂MDH generally have coverage proportions closer to the nominal level than ψ̂MH and ̂NRS in the wide band. However, the differences in coverage proportions were minimal in balanced design (κ = 1.0) among the estimators. Based on the results from the simulation study, I recommend using ψ̂MDH since it generally performs better than the other estimators for all factors considered (bias, absolute bias, MSE, and coverage proportions).
AB - In this paper, the results from a simulation study for the estimation of a common odds ratio in multiple 2 × 2 tables when the data are correlated within clusters are presented. The size of clusters in each stratum is modelled by the negative binomial distribution truncated below 1 with mean cluster size, μ = 4, 5, and 6, and imbalance parameter, κ = 0.6, 0.8, and 1.0. The correlation of the data is modelled by the beta-binomial distribution. A simulation study is conducted to compare the performances of the Mantel-Haenszel estimator (ψ̂MH), the modified Donner-Hauck estimator (ψ̂MDH) and the Rao-Scott estimator (ψNRS, ψPRS) in terms of their biases, absolute biases, mean squared errors, and 95% coverage proportions. The modified Donner-Hauck estimator ψ̂MDH performs better than the other estimators in terms of the bias, absolute bias, and MSE when ψ ≥ 3 in the unbalanced design. In general, there are negligible differences in the bias, absolute bias, and MSE among the estimators when ψ = 1 or κ = 1.0. The estimator ψ̂MDH has generally coverage proportions closer to the nominal level than the other estimators for ρ ≥ 0.3 in the unbalanced design. When κ = 0.8, ψ̂PRS and ψ̂MDH generally have coverage proportions closer to the nominal level than ψ̂MH and ̂NRS in the wide band. However, the differences in coverage proportions were minimal in balanced design (κ = 1.0) among the estimators. Based on the results from the simulation study, I recommend using ψ̂MDH since it generally performs better than the other estimators for all factors considered (bias, absolute bias, MSE, and coverage proportions).
KW - Degree of imbalance
KW - Odds ratio
KW - Simulation
UR - http://www.scopus.com/inward/record.url?scp=0031555797&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0031555797&partnerID=8YFLogxK
U2 - 10.1016/S0167-9473(96)00054-0
DO - 10.1016/S0167-9473(96)00054-0
M3 - Article
AN - SCOPUS:0031555797
SN - 0167-9473
VL - 24
SP - 47
EP - 61
JO - Computational Statistics and Data Analysis
JF - Computational Statistics and Data Analysis
IS - 1
ER -