ABSTRACT
The application of statistical models to biological study, once referred to as biometrics, has a long history. Studies in this field can be based on precisely calibrated laboratory data, or large scale databases that require data analytic methods to be investigated and themselves challenge standard statistical intuition. There is typically biological theory in these settings that may be helpful in guiding the development of likelihood functions and the selection of prior densities. The linear model maximum likelihood model parameters are estimated using least squares and standard errors estimated from the appropriate variance-covariance matrix. Residual plots support the choice of likelihood in the sense that the pattern in the residual plot does not contradict the assumed likelihood. The frequentist analysis is based on the properties of the likelihood, which in the case of normality is essentially least squares, guided by the othogonal decompositions that underlie the analysis of variance testing format.
