Regression with repeated measures in the experimental units.

Abstract

The most satisfactory solution to the problem of modeling a family of regressions with repeated measures in the experimental units is multivariate in nature. However, multivariate methods are difficult to follow and implement. Furthermore, by keeping the focus on the experimental unit, a family of simple univariate linear models will often parallel both the investigator's intuitive grasp of the statistical task at hand. We present two examples based on data from a study of the suckling stimulus during breastfeeding in newborn infants. We show how a family of regression lines can provide useful, if approximate, answers to the questions of interest. One example involves a regression setting proper and the other a typical case of correlation. We discuss alternative univariate models that may be useful for this type of problems.