The general objectives of this research are to describe the small sample problems associated with several of the most frequently used time-series statistical procedures and to develop improved alternative methods. The plan is to provide researchers and applied statisticians with information regarding the negative consequences of employing conventional time-series methods with small sample sizes and to provide more satisfactory procedures for the small sample case. More specifically, the nature and magnitude of small sample bias will be evaluated in a wide variety of descriptive and inferential statistics used in the analysis of simple and complex intervention and non-intervention time-series studies. The investigation will require computer simulations of the sampling distributions of various autoregressive and white noise processes. The focus will be on simple univariate autoregressive models because such models appear to characterize a large proportion of applications in the social sciences. The early stages of the work are expected to specify clearly the weaknesses of several conventional methods. It is anticipated that new procedures developed in the last stage of the research will lead to improved statistical and substantive conclusions in many areas of science that now employ time-series methods. The principal investigators are well qualified to carry out this project. They are expert in both time-series analysis and robustness and they have many years of experience in their respective fields.
