Distance is a fundamental impediment to virtually all economic transactions. Naturally, higher freight costs make trading goods more expensive, and higher migration costs prevent people from moving to other areas where they might have better opportunities of employment. The examples of the importance of distance for economic interactions abound. Perhaps the one area in which many of these arguments do not immediately apply is technology. It is not obvious why technological knowledge cannot be simply put to practice in distant locations at essentially the same cost as in near ones. In the case of technology, one might hypothesise that distance and geography play no role (or a different one, e.g. Caballero and Jaffe 1993).
This argument, although tempting and perhaps partially valid, faces an important hurdle. Countries of the world differ vastly in the level and quality of the technology they use. For example, some plants in advanced economies use techniques, processes, and machines which are essentially unheard of in most less developed countries. This observation suggests that technology also faces significant impediments to diffuse spatially. That is, some friction that resembles transport costs applies to technology diffusion and adoption.
In a recent paper titled “The spatial diffusion of technology” (Comin et al. 2012) we analyse data on the use of 20 major technologies in 161 countries over 140 years to argue that geography, through distance, is an important impediment of technological diffusion. The data come from the CHAT dataset described in Comin and Hobijn (2009) and provides measures of the intensity of the use of technologies such as ATMs, aviation, cars, cellphones, computers, electricity and TVs, to name but a few. We choose important technologies with a wide coverage.
The first step of our analysis describes with simple empirical relationships the temporal and spatial pattern of technology diffusion in data. We find that distance plays an extremely important role in technology diffusion. Being far from technological leaders in a given technology slows down the diffusion of technology significantly. We show that the dependence that we found is not explained by the distribution across countries of income, trade, institutions as measured by political openness, or education as measured by secondary education. The impact of distance on technology diffusion is not proxying for the effect of these factors or of other country-specific factors, but instead seems to reflect that, in fact, distance represents an important obstacle for technology diffusion.
In his bestselling book Guns, Germs and Steel, Jared Diamond hypothesises, and presents some anecdotal evidence, that agricultural technologies diffuse more along the east-west direction than the north-south one, due to their applicability in similar climates. We test Diamond’s hypothesis with our data and find that this is generally the case. Distance is a significantly more important impediment across parallels than across meridians. This finding might be surprising given that most of our technologies are not used in agriculture. Speculating beyond what we can show in our analysis, this might be the result of networks across locations that were created for original agricultural technologies and later have facilitated the diffusion of other technologies in distinct sectors.
So do these findings indicate that technology behaves just as any other good or person that is costly to transport or move? Not quite. Consider a process in which a technology is invented and then diffuses as individuals meet and explain and teach each other about the new technology and how to use it. In such a process, the importance of distance will be governed by the frequency of meetings between agents, as well as by how much more often they meet people that are closer to them. Furthermore, as in any epidemic model, the more agents use a technology, the higher the chances of meeting one of them and therefore adopting it. Eventually, everyone knows and uses the technology. This stylised model of technology diffusion therefore implies that over time, as more people adopt a technology, the importance of distance to the technological leader will diminish, until it eventually becomes irrelevant when everybody has adopted. Thus, this logic suggests that the effect of distance on technology is, in fact, different than the one on goods or people. We should see it disappear over time.
We find very strong supporting evidence for this prediction. The coefficient governing the importance of distance to technological leaders decreases over time (in absolute value since the effect of distance is negative), and tends to converge to a value close to zero in virtually all technologies. Furthermore, when we structurally estimate the simple model described above, we find that it fits the data surprisingly well for 19 out of 20 technologies. (See Figure 1 for two examples.) This indicates that in order to understand the process of technology diffusion, the finding that space becomes less relevant over time is essential.
Figure 1. Effect of distance from technology leaders on diffusion, data (blue) and model (red)
The structural estimation of the model also yields estimates of the frequency of interactions and the decay in the probability of meeting agents that are farther away. As expected, these estimates indicate that newer and network-based technologies are consistent with a higher frequency of meetings. Furthermore, the median estimate indicates that being 1000km farther than the technological leader leads to a 73% decline in the probability of meeting someone that has adopted the technology and can teach us about it. This is a large number that can help explain the technological backwardness of many regions of the world.
Technology is not weightless, at least in the short run!
Caballero, R and A Jaffe (1993), “How High Are the Giants’ Shoulders: An Empirical Assessment of Knowledge Spillovers and Creative Destruction in a Model of Economic Growth”, NBER Macroeconomics Annual, Cambridge. MA: MIT Press.
Comin D, M Dmitriev and E Rossi-Hansberg, 2012, “The Spatial Diffusion of Technology”, CEPR Discussion Paper 9208 and NBER Working Paper 18534.
Comin, D and B Hobijn, 2009, “The CHAT Dataset.", NBER WP 15319.