Do islands trade more or less? A meta-analysis of findings from gravity models

In the geography of trade insularity has been recognized by many as a potentially important factor, but the direction of the impact on trade is unclear. The findings of empirical trade flow modelling are heterogeneous with respect to magnitude, sign and significance of the impact of insularity. While insularity has been a customary included controlling variable in the gravity trade model since the second half of the-1990s, no study exists specifically on the question of the trade impact of being an island. We collected 95 primary studies that deploy the gravity model and use a dummy for islands as a controlling variable in order to meta-analyse the ‘true’ underlying impact in the empirical literature. Following the 2020 MAER-net Protocol for reporting and conducting of Meta-Analysis in Economics, we meta-analyse 95 primary studies that report a total of 2044 estimated island parameters. Our analysis rejects the presence of any (publication) bias in the empirical literature. The primary studies reveal a statistically positive and economically meaningful meta effect. The genuine effect of insularity on trade is significantly positive, also after accounting for the effects of differences in study designs. Our analysis, however, shows that this result also depends on the research methodology, in particular the treatment of zero flows. We also find that the reported effects are larger for primary studies that have been published in peer reviewed studies. Together the identified methodological sources of heterogeneity provide a strong argument for replications of earlier studies with more advanced econometric methods as well as more focussed and dedicated studies on the trade impact of being an island. On balance and in the average, the advantages of being an island (accessibility of sea transports and locations along shipping lines) appear to outweigh the disadvantages (trade cost and geographical discontinuity). Based on the meta-analysis the ‘best practice’ genuine effect conditional on the identified heterogeneity is 0.477 (significant at the 1% level). This finding implies that being an island is on average associated with sixty percent larger trade.