useful for the application of developmental data to policy issues. Of course, suitable confidence bands must be specified for such analysis. The criteria for such confidence bands that I would propose, would at the group level require that there be less than 50% overlap between groups; that is, the mean of the group to be compared would be below the value of the first quartile of the reference group. At the individual level a reasonable confidence band for reliable longitudinal change would seem to be provided by a standard set at one standard error of measurement about the base score (also, see Schaie, 1984b, 1988a).
This presentation has sketched a brief history of my attempts to systematize and improve the methodological rigor with which developmentalists describe changes in behavioral phenomena that occur over age and time. I have consciously restricted my remarks primarily to issues that effect changes and differences in level of function, since several other contributors to this volume deal with the issues of structural change. I have reiterated some of the fundamental aspects of a basic model that describes components of developmental change that have retained value over time. I have also tried to show how the role of developmental studies as quasiexperiments demands certain design complications. As part of such complications I have discussed some of the decisions that one has to make with respect to which threats to the validity of a study must be controlled or estimated and which others can be dealt with by means of specifying relevant and reasonable assumptions. Finally, I have shared with the reader a number of approaches that I have used to present the results of developmentally oriented studies in ways that may make their results relevant to policy issues in a comprehensible and responsible manner. Those readers who attended the 1972 life-span methodology conference or read the publication resulting therefrom ( Nesselroade & Reese, 1973) will probably have noticed that many of the issues I have discussed here are not all that different from those that I tried to address at the earlier conference ( Schaie, 1973). At that time, however, many of my concerns arose from armchair speculation while the prescriptions given in this chapter have at least been seasoned by being tested in a large number of empirical applications. I hope, therefore, that some of the experiences reviewed here will prove useful for the reader's own applications.