analyzing these data exist. As more data regarding age of onset are gathered, the preferred analytic strategy for comparing incidence rates across groups is likely to be survival analysis. Survival analysis is a flexible and powerful statistical method for analyzing incidence of illness when time to onset is known. An interesting aspect of survival analysis is the hazard function, which is the probability of becoming ill at each point in time. Analyses of changes in risk over time may be particularly sensitive indicators of a program' s efficacy and effectiveness. The hope is that the intervention program might begin to exert an effect at its inception and gradually build to its full effect as it is fully implemented with desired impacts on the participants' risk and protective factors. The hazard function curve quantifies the probability per unit time that a participant who has survived up to a particular time will have the onset of the disorder in the very short ensuing time interval. For an insidiously developing disorder, such as schizophrenia, the time to onset may be difficult to ascertain, however, and, at least from the point of view of analyzing a prevention research program, relatively unimportant. Therefore, if survival methods cannot be used, random effects regression models permit the best use of incomplete follow-up data for participants and help avoid some of the problems of sample bias associated with low retention rates during a trial. However, such problems do not disappear; every missing data point or dropout from the study costs some degree of power.
Power calculations should precede the initiation of a preventive intervention trial to determine the requisite sample size. For example, for a universal trial targeting the general population with a short follow-up period to measure the onset of a disorder that has a low baseline frequency and unreliable diagnosis, having one million participants may not yield adequate power to detect statistically significant effects. On the other hand, for a trial of a potent selective preventive intervention sampling a relatively high risk population and using frequent, repeated measurements that are valid and reliable, with a long follow-up period and good retention of subjects, a sample size of 50 per group might be adequate.
When the research program has been completed, the design, sampling, measurement, and analytic decisions should be specified in the peer-reviewed literature and manuals in sufficient detail that they can be replicated by others. The background and rationale are also relevant. When the results have been analyzed, the statistical methods used