In statistics, the ratio of two likelihoods, particularly that of a likelihood function to the maximum likelihood. Widely used as a test statistic, especially for examining relationships among categorical variables arranged in a contingency table. In goodness-of-fit modelling, the smaller the LR, the stronger the relationship, because in using it in this way we attempt to accept a particular model rather than reject a null hypothesis. One of the applications of the LR is evaluating how good a diagnostic test is, and it is claimed that it performs better in this respect than measures of sensitivity and specificity as, for example, it is less affected by changes in the prevalence of a disorder. In this context, the LR is computed by dividing the probability of patients with disease divided by the probability of the same finding in patients without the disease. Now, an LR greater than 1 indicates that the test outcome is associated with the disease while a value less than 1 means the outcome is associated with the absence of the disease. Usually, the LR is converted to a logarithm based on which a significance level (or p-value) can be computed in order to decide whether to reject the null hypothesis in favor of the research hypothesis.
See Diagnosis (or diacrisis), Goodness-of-fit (statistics), Indices of efficacy, Likelihood function, Odds ratio, Prevalence, Sensitivity (epidemiology), Specificity (epidemiology), Surrogate data methods