SP01
Simulating Attrition and Visual Models of Type I and Type II Errors
Haftan M Eckholdt, Albert Einstein College of Medicine

The first example focuses on longitudinal studies as they are plagued by attrition. The second example focuses on studies of new phenomenon that have to history in the literature to help estimate sample size needs and power. While statisticians can addre ss these problems, researchers have trouble understanding the problems. Methods described to help scientists understand the utility of various samples on their hypotheses or impact of attrition on their hypotheses. Starting with the essential hypotheses , the approach begins by characterizing the sample at each stage of research to date: population of origin, recruitment, baseline, first follow up, etc. [BASE & STAT: PROC FREQ, PROC TABULATE, etc,]. The next step executes 1000's of simulations of virtual data or attrition through the random as well as systematic deletion of cases [%MACRO]. In the next step, models are run [STAT: PROC GLM, PROC MIXED, etc.] and relevant coefficients are saved [ODS]. Finally, distributions of coefficients (observed, expe cted, random deletion, and systematic deletion) are compared [GRAPH: PROC GPLOT]. This process will help the scientist to see how their interpretations of hypothesis tests are likely to change with regard to the size and types of samples collected, or the size and types of data lost. This demonstration will include data from several longitudinal government funded studies.