Both academic and practicing accountants use forecasts made by security analysts to estimate market expectations about forthcoming earnings announcements. We report empirical analysis to illustrate how common stock splits induce a loss of precision in computing forecast errors from commonly used analyst forecast data files. An investigation of security price reactions to earnings announcements demonstrates how the loss of precision potentially alters inferences about security price reactions to announcements of earnings that meet, but do not exceed, the consensus forecast. Further analysis indicates that, because stock-split adjustments are made retrospectively, and because firms that execute stock-splits tend to be well-performing ex post, the consequences of the stock-split problem are systematic, potentially contaminating empirical investigations of both time-series and cross-sectional characteristics of forecast errors and of security price reactions to earnings announcements.