This study investigates the difference in average growth rates obtained from two commonly used methods. It is analytically shown that the difference lies on the dichotomy of constant and time-varying growth that can be converted to the dichotomy of trend stationary (TS) and difference stationary (DS) processes. For TS processes the two methods would yield the same results whereas they differ in case of integrated processes. It is also proven that the OLS residuals of a log-linear trend model of an integrated series will be always a random walk, in which case the differenced model that yields the same result as geometric mean is appropriate. The findings are illustrated on the real GDPs of OECD countries.