Abstract
Recent anxieties over the digital divide have centered on the observation that uptake of the Internet is shaped by a number of identifiable, place-based factors. Yet is the Internet any more a product of material geography than previous communication technologies? Our contribution in this article seeks to address this question by deploying quantitative techniques to examine whether the country-level adoption of past communication networks—mail, telegrams, and telephone—was shaped by similar socioeconomic factors. Our results reveal striking similarities in the domestic attributes—income, education, and trade openness—influencing rates of uptake across all four technologies during their major periods of diffusion.
Las ansiedades recientes en torno a la brecha digital están centradas en la observación de que el auge de la Internet está configurado por un número de factores identificables, basados en lugar. Pero, ?‘es que acaso la Internet no es un producto de la geografía material como las anteriores tecnologías de la comunicación? Nuestra contribución en este artículo busca enfrentar esta cuestión mediante el despliegue de técnicas cuantitativas para examinar si la adopción de cadenas de comunicación del pasado a nivel de país—correo, telegramas y teléfono—estuvo determinada por factores socioeconómicos similares. Nuestros resultados revelan asombrosas similitudes en los atributos domésticos—ingreso, educación y apertura comercial—que influyen las tasas de auge en todas las cuatro tecnologías durante los períodos de mayor difusión.
Notes
1Our dependent variable for the Internet is the number of users per 1,000 people. Yet, for all intents and purposes, this measures the same ratio as Internet users per capita.
2Because we are only interested in the determinants of uptake growth, we restrict our analysis to the period before 1970, during which time the number of sent telegrams is expanding in the vast majority of countries. After this date, telegraph usage begins to decline, sometimes quite dramatically.
3For prior discussion of random- and fixed-effects estimators in geographical research, see, for example, CitationJones (1991) and Eriksson, Lindgren, and Malmberg (2008). A more detailed introduction to these estimators can be in found in CitationWooldridge (2009).
4We say “slight” bias because the CitationNickell (1981) bias diminishes as T, the time period covered by the estimations, increases and (with the exception of our Internet estimations) T is large. See Wooldridge (2002) for an excellent discussion of this estimator.
5The coefficient sizes should not be compared with each other across the technologies. The samples are too different, particularly with respect to time, for such a comparison to be useful.
6We have no explanation for why the political constraints variable becomes significantly negative in the case of mail. Yet institutional quality is a variable that changes very little over time. For such variables, it is not uncommon for the estimated coefficient sign to switch in moving from random- to fixed-effects estimations.
7Note that the number of observations is slightly smaller in the GMM estimation compared to the random- and fixed-effects estimations due to the need for instrumenting the lagged dependent and endogenous variable with further lags. The GMM estimation results have to be regarded with some caution as the estimator is more suitable for samples with smaller T. Also, the estimator depends on the assumption that there is no second-order autocorrelation. Fortunately, test results reported in suggest that this hypothesis cannot be rejected.