I show that the median earnings surprise has shifted rightward from small negative (miss analyst estimates by a small amount) to zero (meet analyst estimates exactly) to small positive (beat analyst estimates by a small amount) during the 16 years, 1984 to 1999. I show that a rightward temporal shift in median surprise from negative to positive describes earnings, but neither profits nor losses. Median profit surprise shifts within the positive quadrant, from zero to one cent per share. Median loss surprise shifts within the negative quadrant from extreme negative (about -33 cents per share) to zero. I show that the median surprise for profits exceeds that for losses in every year. I document significant positive temporal trends in both meet and beat analyst estimates for both profits and losses, but I find a greater frequency of profits that either meet or beat analyst estimates in every year. I find a significant positive temporal trend in positive profits that are 'a little bit of good news,' and a significant negative temporal trend in managers who report losses that are an 'extreme amount of bad news.' My results are robust to the four internal validity threats I consider - namely temporal changes in: (1) analyst forecast accuracy, (2) the mix of earnings of one sign preceded by earnings of another sign four quarters ago, (3) the timeliness of the most recent analyst forecast, and (4) the I/B/E/S definition of actual earnings. I find that managers of growth firms are relatively more likely than managers of value firms to report good news profits. I show that when they do report positive profit surprises, managers of growth firms are more likely to report 'a little bit of good news' in every year.