Selection bias in a gravity model of agrifood trade

In the analysis of bilateral trade flows, reported trade of zero or missing observations is quite common and this is a problem when estimating log-linear gravity equations. This has caused many researchers to either ignore the zero trade flows or replace the zero with a small positive number. Both of these actions bias the resulting parameter estimates of the gravity equation. In this study, we correct for this misspecification by using the Heckman selection model to estimate bilateral trade flows for 46 agrifood products, for the period 1990–2000, for 52 countries. In our sample, selection bias rarely affects the signs of variables but often has a substantial effect on the magnitude, statistical significance and economic interpretation of the marginal effects. Hence, treating zero trade flows properly is important from both a statistical and an economics perspective.