We found a match
Your institution may have rights to this item. Sign in to continue.
- Title
Estimating the κ-coefficient from a Selected Sample.
- Authors
Lipsitz, Stuart R; Laird, Nan M; Brennan, Troyen A; Parzen, Michael
- Abstract
Cohen's κ-coefficient (1960) is often used as a measure of agreement between exchangeable pairs of binary ratings on a subject. The coefficient depends on the marginal probability of a positive rating. Suppose that the probability of a positive rating in the observed sample is not applicable to the group of interest. We estimate the (2 x 2)-table of probabilities for a pair of raters from the observed sample. Assuming that the odds ratios in the observed sample and the group of interest are the same, we use iterative proportional fitting to rescale the (2 x 2)-table to have marginal probabilities equal to those from the group of interest. Finally, we re-estimate the κcoefficient from this rescaled table. Our example is from the Harvard medical practice study in which the 'rating' is a physician's view of whether an 'adverse event' occurred in the hospital; an adverse event is an unintended injury caused by medical workers in a hospital rather than the disease process. The group of interest is the whole population of hospital patients, but the observed sample consists of patients who are at high risk for an adverse event.
- Subjects
STATISTICAL sampling; PROBABILITY theory; ESTIMATION theory
- Publication
Journal of the Royal Statistical Society: Series D (The Statistician), 2001, Vol 50, Issue 4, p407
- ISSN
0039-0526
- Publication type
Article
- DOI
10.1111/1467-9884.00286