Economics of Technology

Abstract

This thesis investigates the spread and activity of the Internet over the years 2006-2012, an unprecedented period of growth in global internet connectivity. In stark contrast to previous general purpose technologies such as steam power or electrification the internet's diffusion process operates at a much faster pace. Whereas it took 100 years for the steam engine to reach saturation, and around 60 years for electrification, we estimate that the average time to saturation for the internet's expansion has been just 16 years (1% to 99%). Significantly, whilst the emergence of the steam engine necessitated the co-development of the engineering profession to install, use and maintain the engines, every participant in the Internet obtains (and extends) the knowledge required to utilise the internet for greater productivity. Moreover, each internet user, simply by their pattern of interaction with the internet reveals personal preferences and choices, and this, combined with the global reach and remarkably democratic foundations of the internet enables social science research of an unparalleled nature. I present a novel data set with corresponding methods to analyse the diffusion of the internet at a very disaggregated level and show it's relationship with economic progress, structural changes and innovation. The first two chapters of this thesis focuses on the extensive methods necessary to aggregate a trillion random Internet (IPv4, ICMP) probes and transform this information into useful data. This work yielded novel data join algorithms to optimise the joining process with distributed data flows. The resultant data set comprises accurately geo-located IP activity ('online'/'offline') aggregated at over 1600 locations (cities) at 15min intervals, over the period 2006-2012. The third chapter applies the data set to investigate biological, social and economic behaviour. The American Time Use Survey (ATUS), in combination with our Internet data, is used to upscale the time diary to cities worldwide. A novel machine learning method is presented to facilitate the process of predicting sleep start and sleep stop time over seven years around the globe. Furthermore, the dynamics of the Internet Protocol space regarding allocation challenges are overcome to derive comparable monthly Internet per capita measures and estimate country-specific diffusion properties. The globally consistent regional estimates are used to research the effect of the spread of the Internet on local economic outcomes. I find that the propagation of the Internet had positive effects on the economic performance of regions across industries. On the other hand, in industries where the spread of the Internet has lead to the possibility to outsource operation and results in a structural change, the spread of the Internet has adverse economic effects. The final chapter investigates how the free distribution of information over the Internet is seen as a threat by autocratic governments. First a method to identify events of Internet censorship through speed tampering is developed, and validated for Iran. Then, the method is applied to identify the limiting of the market for information during the Russian presidential election in 2012 using a difference in difference design. I find that on average the government candidate achieved a lower-bound estimate of 3.2 percentage point increase on Election Day due to subtle tampering of the Internet.