AbstractOver the last 5 years, the top accounting journals have published 388 articles that incorporate an empirical proxy for accruals‐based earnings management. Researchers use these proxies to measure diverse managerial activities that represent fundamentally different constructs (from beneficial earnings management at one end of the spectrum to earnings manipulation at the other). We present a simple framework that defines the specific construct of “earnings manipulation” and places it within the context of the broader concept of “earnings management.” At the construct level, one distinguishing characteristic of earnings manipulation is that accruals reversals cause future earnings to be lower than earnings would have been absent the manipulation. We explore the extent to which increases in 11 common earnings management proxies—that account for 88% of the research published in the last 5 years—result in future earnings that are lower than the firm's average earnings to determine which measures are more likely to reflect earnings manipulation. Our results indicate substantial variation across earnings management proxies in this characteristic with some (including the most commonly used proxies) consistently exhibiting a positive association with future performance, suggesting that they generally do not reflect earnings manipulation. Supplemental analyses that consider earnings persistence and performance of the proxies across firm life‐cycle stages yield similar inferences. Finally, we propose two simple empirical solutions that researchers can implement to capture the construct of earnings manipulation.