This study proposes the return-earnings covariance as a proxy for the usefulness of earnings inferred from the absolute magnitude of price changes associated with earnings information. It is argued that such measurement of the absolute usefulness of earnings information has been neglected in existing long-window studies. For example, the ERC and R² measure the marginal impact and the relative impact of earnings information on the stock price, respectively. It is demonstrated that the return-earnings covariance is a close proxy for the absolute magnitude of price change which is free from noise in both return and earnings. Thus, the return-earnings covariance can be used in long-window studies as well as short-window studies. Two covariance measures, the total covariance and the time distribution of weekly covariance are introduced and applied to empirical data to show new insights that can be obtained by the measures. The result indicates that the previously documented decrease on the value relevance of earnings over the past decades is mainly driven by the increasing influence of factors not directly related to earnings on the regression measures, not by a decrease in the absolute usefulness or timeliness of earnings. It is also found that the previously documented weak return-earnings relation over the short-window announcement period or contemporaneous return-earnings association for larger or more closely followed firms is due not only to more vigorous pre-disclosure information production activities of those firms, but also, and more importantly, to the weaker overall magnitude of price changes associated with earnings information of those firms.