This event is hosted by the Department of Biostatistics.
PRESENTER: Jeffrey Simonoff, PhD Professor of Statistics NYU Stern School of Business
ABSTRACT: In this talk I highlight a very old idea that has become very timely in recent years. I describe effect codings as an alternative quantitative practice for analyzing and interpreting categorical predictors in regression models for applied research. Effect codings highlight comparisons against an overall level as a means for interpreting information. This technique is appropriate in any research context where a categorical predictor does not have a natural reference group, but a particularly important application is to racial groups. This procedure, in contrast to the use of indicator (dummy) variables, does not position responses of any particular raced group as normative — the standard against which all other race effects should be interpreted. I also discuss how effect codings can be easily extended to situations where subjects can be members of multiple groups (such as being multi-racial), taking advantage of another very old idea, that of regression models with linear restrictions imposed on the parameters.
*Note: this should not be considered as a technical research talk; non-technical and student audiences are encouraged to attend.