772 BIRKEBrynjolfsson andKemerer (1996)Software US market data fromup to 22 spreadsheetprograms between1987 and 1992Hedonic price modelexplaining price as afunction of networksize and productcompatibility withdominant standardSignificant impact of networkeffects on price: a 1% increasein the product’s installed basewas associated with a 0.75%increase in price. Productswhich adhered to thedominant standardcommanded over a 46%higher priceGuptaet al.(1999) Digital TV 500 randomly selectedrespondents fromChicago and NewJerseyModel ofmarket-mediatedinterdependencebetween hardwareand softwareproviders; inclusionof marketing mixeffects; productdemonstration andquestionnaire tocalibrate modelInterdependence betweenhardware and softwareproviders plays an importantrole in acceptance of HDTV.Neglecting this interactionwould bias in favour of HDTVGandalet al.(2000) CDs US market data on CDplayers and CDs inthe years 1985–1992Structural model of theCD player marketwith hardware andsoftware providersis estimated usingmarket data;simulation ofcounterfactualsA 10% increase in CD titles hasthe same effect on CD playersales as a 5% price cut in CDplayer prices. Downwardcompatibility is estimated tospeed up the diffusion processby 1.5 years

THE ECONOMICS OF NETWORKS 771Table 1.Empirical Studies on Network Effects.Study Industry Data Empirical approach Main findingsCabral and Leite(1992)Telex Annual data on thePortuguese Telexmarket from 1962 to1987Reduced-formequations explainingusage and accessA 10% increase in network sizeimplies a 1.6% increase inaverage consumption, a 0.7%short-run increase and a15.9% long-run increase inthe equilibrium number ofsubscribersShurmer (1993) Packaged PC software 270 UK PC softwareusersOrdered probit modelto gauge theimportance ofdifferent sources ofnetwork externalitiesNetwork effects for PC softwarehave different sources(add-ons, books, training, etc.)and the importance of thesesources vary by user typeGandal (1994) Spreadsheet software Annual data on USspreadsheet softwareduring 1986–1991Estimation of hedonicprices with networksize as one of theindependentvariablesConsumers are willing to pay aprice premium for softwarecompatible with market leaderLotus 1-2-3Saloner and Shepard(1995)ATMs 4500 US commercialbanks during1972–1979Survival analysismodel of productadoptionBoth network effects andeconomies of scale arepresent in the market. Addinga branch in limited branchingstates increased the hazardrate by 6% and added 0.9% tothe overall adoptionprobabilityEconomides andHimmelberg (1995)Fax machines Fax machines sold inthe USA between1979 and 1992Structural model ofdemand is calibratedwith market dataCalibration suggests thatdemand growth was stronglyinfluenced by networkexternalitie

n the era of the platform, the future remains open. Answers to crucial questions are for the moment unknowable. The answers depend on our choices, not just on the technology. For example, will cloud technologies and the platform-driven economic reorganization they cause drive the productivity growth on which sustained real income improvement occurs? Will these reorganizations destroy jobs or reduce the required skill levels?The technologies—the cloud, big data, algo-rithms, and platforms—will not dictate our future. How we deploy and use these technologies will. When we look at the history of innovations such as electric utility grids,call centers, and the adoption of technology standards, we find that the market and social outcomes of using new technologies vary across countries. Once we start on a technology path, it frames our choices, but the technology does not determine in the first place exactly which trajectory we will follow.We will be making choices in an inherently fluid and ever-changing environment shaped to some degree by unpredictable technical change and social reactions to these changes. Ultimately, the results will depend on how we believe markets should be struc-tured—who gains and who can compete; how we innovate; what we value in society; how we protect our communities, our workers, and the clients and users of these technologies; and how we channel the enormous opportunities created by these sociotech-nical changes. It is up to us to sidestep a dystopia and to create, if not a utopia, at least a world of ever greater benefit for communities and citizens.

His dystopian vision is now finding full expression in the fear that digital machines, artificial intelligence, robots, and the like will displace work for a vast swath of the population.

However, despite the potential beneficial benefits of the gig-economyfor the workers' welfare, also in terms of flexibility, these aspects should notbe overestimated. Whilst it is certainly true that most jobs in the gig-economycome with a flexible schedule, this does not say really much on the overallsustainability of these arrangements: competition between workers, that insome cases is extended on a global dimension through the internet,33 pushescompensations so down that people may be forced to work very long hoursand to give up a good deal of flexibility in order to make actual earnings.

In Coasianterms, they facilitate a further reshaping of "market" and "hierarchy"patterns,17 in addition to the already known "fissured workplace"'18 and"hierarchical outsourcing" discourses. 19 In fact, both crowdwork and "work-on-demand via app" allow for a far-reaching "personal outsourcing" ofactivities to individuals rather than to "complex businesses." This, as it willbe shown below, grants even more leverage to standardizing terms andconditions of contracting out and assigning work whilst keeping aconsiderable control of business processes and outputs.

This, in general,allows minimizing transaction costs and reducing frictions on markets. Therapidity within which job opportunities are offered and accepted and the greataccessibility to platforms and apps for workers makes it possible to accede tovast pools of people available to complete tasks or execute chores in a precisemoment of time.

These forms of work, of course, present some maj or differences amongeach other, the more obvious being that the first is chiefly executed onlineand principally allows platform, clients, and workers to operate anywhere inthe world, whilst the latter only matches online supply and demand ofactivities that are later executed locally. An obvious consequence is that thismatching can only occur on a much more local basis than what happens withcrowdwork

In "work-on-demand via app," jobs related to traditional workingactivities such as transport, cleaning and running errands, but also forms ofclerical work, are offered and assigned through mobile apps.

The nature of the tasks performed oncrowdwork platforms may vary considerably. Very often it involves"microtasks": extremely parceled activities, often menial and monotonous,which still require some sort of judgement beyond the understanding ofartificial intelligence (e.g., tagging photos, valuing emotions, or theappropriateness of a site or text, completing surveys)