How Technology Spreads

07/01/2001
Summary of working paper 8130
Featured in print Digest

Computer adoption strongly depends on having high levels of education of the labor force.

The adoption of new technology is essential to long-term macroeconomic growth. In general, rich countries are on the technology frontier and rely on research and development to achieve further improvements in technical efficiency. Low-income countries, in contrast, have the option of adopting technologies already developed elsewhere. Yet not much is known about the process by which new technologies spread from one country to the others.

In Cross-Country Technology Diffusion: The Case of Computers (NBER Working Paper No. 8130), co-authors Francesco Caselli and Wilbur John Coleman II use cross-country panel data on computer imports from 1970-90 to analyze the determinants of technology diffusion. The idea is that for the many countries that do not have a domestic computer industry, computer imports are a measure of the flow of new computers installed in the country, and are therefore a good proxy for computer adoption. The authors use three different datasets based on United Nations trade and production data: the first sample uses computer import data for all the countries with available information; the second excludes from that data the countries that report positive computer exports. The third sample estimates a proxy variable for computer adoption equal to computer production plus imports minus exports.

Caselli and Coleman find that computer adoption strongly depends on having high levels of education of the labor force, a result that supports the view that there is a skill-bias in technology adoption (at least for computers). Another important determinant is the source and type of trade with other countries: countries with large manufacturing imports from OECD countries adopt computer technology more readily. Property rights protection, high investment per worker, a small share of government and agriculture in GDP, and a large share of manufacturing in GDP, are other variables that seem to accelerate computer adoption. In contrast, the share of the population that speaks English does not have a significant effect.

The authors use per-capita income and regional "dummy variables" as proxies for other unknown variables omitted by their model. Since these proxies are sometimes significant, the authors conclude that there still are undiscovered determinants of computer adoption that remain to be found.

-- Noshua Watson