Confidence of 100 per cent sensitivity and specificity
Valenzuela, C.Y.
Revista Medica de Chile 125(2): 228-235
1997
ISSN/ISBN: 0034-9887 PMID: 9430946 Document Number: 474536
A new strategy and four new methods are presented to calculate the limits of the confident interval for an estimate of a proportion equal to 1.0 or 0.0. A current formula which includes 1/(2n) for continuity correction leads to a confident interval which does not include the parameter estimate. Thus, it is proposed: 1) The exclusion of the factor 1/(2n) in that formula leads to correct most of its inconsistencies. The new strategy assumes that the upper limit of a confident interval when the estimates is 100%. The lower limit is calculated by assuming that there is a proportion in the population, from where the sample was taken, such as the probability of getting 100% in the sample is equal to the probability of falling into type I error of current statistics (0.05, 0.01, etc). Three methods are proposed with this strategy. 2) A combinatorial solution based in the knowledge of the number of individuals at whom the test can be applied. 3) A solution based on the binomial distribution. 4) A solution based on the Poisson distribution.