Applied Statistics
International Conference, 19 - 22 September 2004
Ljubljana, Slovenia
Statistical Principles to Live By
Frank E Harrell Jr
Department of Biostatistics
Vanderbilt University School of Medicine Nashville, Tennessee
USA
This talk deals with principles derived from over 30 years of applying
statistics to biomedical research, collaborating with clinical and basic
biological researchers and epidemiologists. The principles relate to
statistical efficiency, bias, validity, robustness, interpretation of
statistical results, multivariable predictive modeling, statistical
computing, and graphical presentation of information. Topics to be discussed
include respecting continuous variables, avoiding non-descriptive
statistics, problems associated with filtering out negative results,
overfitting, shrinkage, adjusting P-values for multiple comparisons without
adjusting point estimates for same, and the false promise of multi-stage
estimation and testing procedures, related to the use of bogus conditional
techniques for computing what is advertised as unconditional variances or
type I errors.
|