|
Abstract:
|
In many statistics problems there is no true model for the data and no meaning to the probability of rejecting a true hypothesis, the accuracy of estimating a true value, or the probability with which a confidence procedure covers a true value. Instead of trying to estimate or cover a true value, we should compare models and their parameters according to how well they describe the data at hand. Likelihood is the primary way to quantify the quality of a probability model for describing data and is the main statistical concept we should teach. This talk provides several illustrative examples.
|
