Abstract
This meta-analysis summarizes and explains the variation in the deterring effect of distance on tourism flows by analyzing 870 estimates from 138 primary studies utilizing data covering the last 25 years. We find substantial heterogeneity among studies that mostly correlates with (unobserved) study characteristics, estimation methods, and locations of origin and destination included in the primary studies. We confirm previous findings that the mean total distance decay effect, using preferred methods and datastructures, is close to unit elasticity in absolute value (-0.93). However, when controlling for mediator variables, we find that the direct, physical, distance decay effect is significantly lower (-0.74). This distance decay effect is remarkably stable over the last 25 years and reveals a positive relation between distance and the total number of tourists.
Graphical abstract
Main findings
This meta-analysis has some intriguing, well, in our eyes, elements. First, we provide a micro-economics framework for tourism flows (although only online accessible in the online appendix). Second, we argue that additional covariates in our meta-analysis, such as controlling for islands and world heritage sites, act as mediators. Again, in our online appendix, we provide a causal framework (and the needed assumptions) for this. See the two directed acyclical graphs in Figure 2.
First, we specifically state the assumptions needed for a mediating framework. Adopting the general framework of Pearl (2009) three possible mechanisms of confounding for causal identification may arise: mediation bias, unobserved heterogeneity bias, and collider bias. We argue in Figure 1 in the paper for the presence of mediation bias. Controlling for possible unobserved heterogeneity bias, as often is given as the rationale in meta-analyses to include other variables, is presented in the left panel of Figure 2. The right panel of Figure 2 depicts the general presence of unobservables where especially the unobserved confounding factor U is problematic as it leads to collider bias as controlling for, in this case, the variable ‘Island’ opens up the ‘path’ for U (see for an extensive treatment Celli 2022).
In the paper we assume that distance itself is not influenced by other factors—arguably, a weak assumption. This implies that the setting in the left panel of Figure 2 is not applicable for our case and unobservables (V, W) do not play a role in the right panel of Figure 2. Unfortunately, we cannot rule out the possible impact of U. For a possible claim of a mediation framework we have to make the assumption that U does not impact the mediation variable (or \(\text{U} \perp \text{`Island'}\)). For the variable ‘Island’ and when controlling for regional variables, as we do, this is not a strong assumption. However, for more socio-economic variables, such as ‘Exchange rate’ and ‘Regional trade agreement’ these are much stronger assumptions. Note, that our empirical results align with these assumptions. Namely, adding more variables does not increase out-of-sample model performance.
A second finding is that the distance decay does not seem to change over time, neither for international of domestic tourism, which to a certain extent contradicts Rosselló-Nadal and Santana-Gallego (2024).
Third and finally, as can be seen in our graphical abstract Figure 1, the fact that we find an inelastic distance decay states that there is a positive relation between distance and the total number of tourists (so with constant distance decays over times). With rising income levels across the world, this might serious (environmental) consequences