The paper addresses the issue of interpreting parameters in log-linearized models as elasticities, which can be misleading in the presence of heteroskedasticity. It argues that the standard practice of using ordinary least squares (OLS) to estimate log-linearized models, such as the gravity equation for trade, is flawed because it violates the assumption that the error term is independent of the regressors. This leads to biased estimates of elasticities. The paper proposes a pseudo-maximum likelihood (PML) estimator as a more appropriate method, which is consistent in the presence of heteroskedasticity and can handle zero values of the dependent variable. The gravity equation, a key model in international trade, is analyzed using both the traditional specification and the Anderson-van Wincoop specification, which includes fixed effects. The paper finds significant differences between estimates obtained using the PML estimator and those obtained using OLS. These differences persist even when the gravity equation accounts for multilateral resistance terms or fixed effects. The PML estimator provides more accurate estimates of the determinants of trade, such as GDP, distance, and colonial ties, and shows that the roles of these variables differ significantly from those predicted by the log-linear tradition. The paper also discusses the importance of addressing heteroskedasticity in econometric models, particularly in constant-elasticity models, and highlights the limitations of the log-linear approach. It argues that the presence of heteroskedasticity can lead to strikingly different estimates when the gravity equation is log-linearized rather than estimated in levels. The paper concludes that the PML estimator is a more robust and reliable method for estimating gravity equations and other constant-elasticity models, as it accounts for heteroskedasticity and provides more accurate estimates of the parameters of interest.The paper addresses the issue of interpreting parameters in log-linearized models as elasticities, which can be misleading in the presence of heteroskedasticity. It argues that the standard practice of using ordinary least squares (OLS) to estimate log-linearized models, such as the gravity equation for trade, is flawed because it violates the assumption that the error term is independent of the regressors. This leads to biased estimates of elasticities. The paper proposes a pseudo-maximum likelihood (PML) estimator as a more appropriate method, which is consistent in the presence of heteroskedasticity and can handle zero values of the dependent variable. The gravity equation, a key model in international trade, is analyzed using both the traditional specification and the Anderson-van Wincoop specification, which includes fixed effects. The paper finds significant differences between estimates obtained using the PML estimator and those obtained using OLS. These differences persist even when the gravity equation accounts for multilateral resistance terms or fixed effects. The PML estimator provides more accurate estimates of the determinants of trade, such as GDP, distance, and colonial ties, and shows that the roles of these variables differ significantly from those predicted by the log-linear tradition. The paper also discusses the importance of addressing heteroskedasticity in econometric models, particularly in constant-elasticity models, and highlights the limitations of the log-linear approach. It argues that the presence of heteroskedasticity can lead to strikingly different estimates when the gravity equation is log-linearized rather than estimated in levels. The paper concludes that the PML estimator is a more robust and reliable method for estimating gravity equations and other constant-elasticity models, as it accounts for heteroskedasticity and provides more accurate estimates of the parameters of interest.