Abstract
Celebrities are anecdotally one of the most observed groups in contemporary society, but are difficult to capture in large-scale quantitative empirical analyses. In this analysis, we use a unique dataset, Getty Images photographs, to study this social group and its various network structures. Overall, our analysis demonstrates that celebrity networks are characterised by low degrees of separation and high connectivity between one another. Higher industry status (‘talent’) and media profile form particular, more exclusive, networks within the larger universe of celebrities. Our empirical results stack up with the theory of ‘superstars’ and the ‘rich-get-richer’ model of preferential attachment and cumulative advantage. We speculate that there may be substantial social and economic outcomes to being more connected to other celebrities that transcend the social ties observed at the documented social events.
Notes
1. 2009 is the first year this data is available. Please see a more comprehensive description of Forbes’ methodology: http://www.forbes.com/2009/02/06/forbes-star-currency-methodology-business-media-star-currency-09_0210_methodology.html While there are a number of measures of ‘talent’ (awards won, number of film credits and so on), the perceived ability to draw audiences, other top stars and resources to the film's production is as good a measure as any in Hollywood show business.
2. The bot is written in the programming language Perl and uses software agents to gather the web pages and parser to break the web page into structured information. The meta information is stored in a relational database (MS-SQL server).
3. We constructed and analysed these networks using a number of different tools and programs. For example, using the data-mining tool SPPS Clementine we were able to identify groups in the network. The social-network analysis (SNA) tools Pajek and Netminer were utilised for the analysis.
4. shows the degree distribution of the whole network on a log-log axis. Plotting power distribution on a log-log axis yields a straight line. Statistically the network conforms with the power distribution (R-square 0.867 with significance 0.00) with Frequency = 91883.2*Degree^(–1.586). ANOVA testing indicates that the fit is significant. An eye examination of the log-log graph () indicates that the frequency of the lower part of degrees is not high enough. It means that our results would be strengthened if we had more low-degree participations. Those with low degree are people that attend only small private events which usually are not photographed.
5. Small-world networks are characterised by measuring the Clustering Coefficient (CC) and the Characteristic Path Length (CPL) and comparing these results to a random network (Bornholdt and Schuster Citation2003, Wasserman and Faust 1994). The CPL measures how many liaisons (or other nodes, that is, degrees of separation) there are between any two nodes within a network. The CC measures how many of a node's contacts are connected to each other (see Uzzi et al. Citation2007 for a terrific and thorough review of small-world network literature). While most random networks have a short CPL (Watts and Strogatz Citation1998) the CC of random networks is quite low. Small-worlds, however, have both short CPL and very high CC compared to a random network. In our network n, number of nodes, equals 6754, k, average degree equals 0.1605.The requirement stated by the hypothesis is therefore that Q >1. where Q is the ratio between CPL ratio and CC ratio (Uzzi et al. Citation2007). In our network Q = 3.71 (Watts and Strogatz, Citation1998, Grossman, 2002).
6. For the mathematically inclined, please see the Appendices for social-network analysis corresponding to our general results.
7. We cross-referenced our results using another similar ranking of industry status, Neilson Media Star Power index. Our results were almost identical to the Forbes Star Currency ranking.