ABSTRACT
With the fixed-effects model, the conclusions apply only to the treatment levels tested. With the random-effects model, the treatment levels tested are only a sample of all the possible levels that we wish to consider; the conclusions then apply to all these possible levels. The most common differentiation between fixed and random effects concerns human subjects. If human subjects are being tested as a sample of the population, the conclusions drawn from these subjects will be extended to the whole population. Subjects are a random-effects factor. Sometimes the conclusions drawn from the subjects are not extended beyond the subjects actually tested; in this case the subjects are a fixed-effects factor. The difference between fixed-effects, random-effects, and mixed-effects models merely involves a difference in the final step of the ANOVA; it involves merely the use of an interaction mean square for some of the denominators for the F values, rather than error mean squares.
