The diffusion of innovations has always represented a fruitful topic of research in the field of innovation studies. Scholars in this area have discussed at length about the most relevant factors influencing the timing of adoption and alternative explanations of the determinants of the diffusion pattern have been proposed. A first class of models focussed on the role of information in explaining the diffusion of a new product. This is the case of epidemic models (Griliches, 1957; Mansfield, 1961), in which 'word-of-mouth' is the only mechanism to get to know about a new product or technology and, as a consequence, the innovation spreads among its potential adopters like an infectious disease by personal contact. A second class of models focussed on the characteristics of individual adopters as determinants of adoption and claimed that diffusion is not instantaneous because adopters are heterogeneous. These are Probit (or 'rank') models (David, 1966, 1969, Davies, 1979) in which rational agents are assumed to be fully informed about the innovation. A third class of models examines the effect of the stock of previous adopters on the decision of potential adopters. This effect can be negative, as in the case of the so-called 'game theoretic' models, in which an increase in the number of adopters negatively influences its profitability in the future (Reiganum, 1981), or positive, as in the case of models in which the stock of previous adopters is meant to produce a positive externality. This type of externality has been conceptualised in different ways, for example as a direct or indirect network effect (Katz and Shapiro, 1985, 1994; Farrel and Saloner, 1985), as an informational cascade (Bikhchandani et al., 1992 and 1998; Banerjee, 1992). Finally, a last class of models claims that returns on adoption depends on the position of a particular adopter in the order of adoption. In particular, early adopters are supposed to gain a sort of first-mover advantage that later adopter will not be able to exploit (Ireland and Stoneman, 1985; Fudenberg and Tirole, 1985). Late adopters are subject to two opposing forces. The first one is a bandwagon effect triggered by the observation of previous adopters and resulting in phenomena such as fads and fashions. The other one is a 'snob effect' which can occur when potential adopters reject the innovation trying to look different than others (Abrahamson and Rosenkopf, 1993). Empirically, few papers have provided a comparison across different models (the notable exceptions being Karshenas and Stoneman, 1993; and Zettelmeyer and Stoneman, 1993). The aim of the present paper is to study empirically the impact of each of the above-discussed factors on the adoption and diffusion of a consumer technology. In particular, we will test the alternative models in the context of the diffusion of portable Digital Audio Players (DAPs), one of the most innovative and successful hitech product in the last decade. Our empirical analysis relies upon a survey of 1562 young potential adopters from 8 European countries (France, Germany, Italy, Portugal, Netherlands, Spain, Switzerland, UK) and Japan. We will perform duration analysis to understand the factors that affect the conditional probability that a user adopts a DAP at a time t, given that he or she has not adopted at t-1. Both semi-parametric and parametric models will be estimated including different specifications of the baseline hazard function. This methodology and the type of data collected will allow us to test whether the assumptions underlying the most common models of diffusion provide an adequate explanation of the pattern of diffusion of DAPs for the users in our sample. The paper is structured as follows. Section 2 provides a review of the literature on innovation diffusion, presenting the models that will be tested on our data. Section 3 describes the market of digital audio players. Section 4 constitutes the core of our analysis. It first presents the data used for the analysis and then illustrates the model and the variables used to test the effect of different factors on the conditional probability of adoption. Section 5 illustrates the results of our estimates and Section 6 concludes
