STATISTICAL METHODS IN HUMAN DEVELOPMENT/CLINICAL STUDIES

点击翻译按钮获取中文摘要

This project seeks to develop new statistical techniques for problems in human development and to apply existing techniques to those problems in novel ways. Work continues in three areas: (1) understanding the birth weight distribution in human populations, (2) developing methods to test whether log link or the logit link is better suited to a given data set, (3) understanding age, period, and cohort effects on cancer incidence. We have recently elucidated a process of round-off, conversion from ounces to grams, and grouping into weight classes that induces an artifactually jagged frequency polygon in United States birth weight distributions displayed in 100-g bins compared to those from countries where the original weight units are grams. We are developing a one-parameter family of link functions for modeling binary outcomes that allow the data analyst to use likelihood ratio statistics to examine whether a log link or a logit link better suits a given data set. Age-period-cohort models are commonly used to examine changes through time in cancer incidence. We will apply these models to analyze the recent update of the SEER data and examine how to interpret such analyses in light of the confounding among linear trends in age, period, and cohort effects inherent in the model formulation.