Features as predictors of phone popularity: An analysis of trends and structural breaks

This article analyzes dynamic changes in mobile phone popularity based on phone features. The time period, 2004-2013, selected for the study is interesting because many technical innovations took place molding the mobile phone market dramatically. The study utilizes comprehensive phone model and sales data collected in Finland, combined with temporally ordered probabilistic models to discover the time behavior of predictivity. More precisely, the Tree Augmented Naïve Bayes - classification method is adapted to detect those phone characteristics that best predict the annual phone popularity measured as phone model unit sales. Linear regression and the Chow test are used to discover potential trends and structural breaks. The strength of the predictivity is measured as Kullback-Leibler Divergence. This kind of systematic longitudinal analysis highlights patterns, which are otherwise not possible to observe. The study discovered that the operating system is clearly the only feature with an increasing strength in predicting popularity over time. In contrast, sixteen features have structural breaks between 2004 and 2013. Most such breaks are related to the technical evolution of phones: their display, communication, and camera capabilities. Notably, the structural break in 2007-2008 related to the phone manufacturer brand is interpreted as the market turning from hardware to software driven mode, which contributed to Nokia's failure with Symbian and Windows operating systems, and to Google's success with a hardware independent operating system Android.