逆抽样模型下流行病学优比统计推断.pdf

Æ Æ Æ Æ Æ Æ Æ ÆÆ Æ ÆÆ Æ Æ ÆÆ Æ Æ Æ ÆÆ Æ Æ Æ Æ Æ Æ Æ ÆÆ  Æ Fieller’ Delta F ÆÆ Æ  Æ Æ Æ 95% ÆÆ R ÆÆ Æ ÆÆÆÆ ABSTRACT The statistical inference of epidemiological risk indices has always played an important part in epidemiological investigations. The frequently used appealing indices to measure the extent of association between a risk factor and an outcome include risk difference, relative difference, odds ratio and generalized odds ratio, etc. Each index has its own advantage and disadvantage in terms of interpretation and statistical properties. However, in applications, one of the more serious problems facing us when we using these indices is that the underlying disease is rare, so the probability of obtaining only a few or zero cases in a sample under binomial sampling can be large or non-negligible. For having advantage in the situation, we employ inverse sampling. In addition, although existing asymptotic methods under inverse sampling provide statisticians and epidemiologists with a fast device in large-sample studies, their reliability in small-sample studies has long been suspected and criticized. To get out of the awkward predicament, we’ll employ saddlepoint approximations which are gradually emerging as powerful tools for dealing with small sample size and sparse data structure. And no surprise to see that during the past two decades, research in the area has vastly increased. The main objective of this paper is to derive three confidence intervals for odds ratio in inverse sampling. They are derived on the basis of the score statis- tic, likelihood ratio statistic and the saddlepoint approximations approach, re- spectively. To c