Lawrence Berkeley National Laboratory

Simple statistical models are used to illustrate two important issues arising in the analysis of gouped data. the consequences are explored of grouping continuous data and analyzing the resulting contingency table. Specifically, an expression for the loss of power is derived when an odds ratio is used to assess risk measured by a continuous variable. Also explored are the consequences of employing correlation and regression coefficients to analyze summary variables derived from grouped data (ecologic data). An expression is given that demonstrates the magnitude of a bias (ecologic fallacy) resulting from analyzing a specific type of grouped data.