Missing data lead to unreliability of counts and rates, thereby reducing the effectiveness of statistical techniques used in causal analysis. Because data are unlikely to be missing randomly, it is possible that time-series trends will be portrayed incorrectly and that coefficients in linear models will be biased. This paper outlines U.S. crude homicide trends that emerge from direct use of published death registration notices, and discusses specific missing data problems with earlier data sets. Regression- based econometric forecasting is used to derive year-by-year forecasts of U.S. crude homicide rates; these are compared with previous estimates. Finally, limitations of this methodology are described. 1 table, 3 figures, and 25 references