The impact of web traffic on revenues of traditional newspaper publishers in total traffic would lead to a 6.41% decrease in total revenue. Consequently, a 100% decrease in referral traffic would imply a 4.2% decrease in total revenues. The coefficient of the variable log(GDP per capita) implies that the variable has a positive impact on publishers’ revenues. As the citizens of a country grow wealthier, publishers may attain higher revenues. The countries examined also exhibit different GDP per capita patterns over the period. Negative coefficients on the variables 2012 and 2013 indicate a declining trend in newspaper publishers’ revenues over time. This finding is consistent with the general market trends discussed in the main body of the report. The time period of three years does not allow for significant year-on-year variations in some of the other key variables estimated by the fixed effects model. As a result, the properties of the fixed effects model do not identify a significant impact of circulation and paywall on revenue. However, an alternative approach undertaken using a random effects model does capture a positive and significant effect of circulation, even when controlling for other firm-specific factors, while also maintaining a 60 similar, significant coefficient on log(total traffic). Given the advantages of estimating a fixed effects model in this study, the random effects model was not used in this report.61 Finally, the model has not identified a statistically significant relationship between advertising spending, as measured by log(advertising spending), and revenues for these publishers in the sample over this time period for similar reasons as for circulation. 5.1.5 Robustness checks To assess the sensitivity of the estimates, additional specifications were run to test how the coefficients of interest varied if other variables were excluded. These specification tests did not produce material differences in the coefficient on log(total traffic) and did not affect its significance. Recent literature by Cozzolino and Giarratana (2014) suggests that endogeneity may have been 62 introduced with the inclusion of both circulation and total traffic variables in the model. Additional specifications omitting circulation as well as employing a two-stage least squares estimation using the instruments proposed by Cozzolino and Giarratana have been estimated. Differences between the coefficients of these specifications and the original specification were not material. 5.1.6 Calculating the value of a visit The average value was derived from the estimated revenue impact on the publishers in the sample and the traffic they received as follows: Estimated revenue impact / total traffic = average value of a visit The estimated value of a visit ranges between €0.04 and €0.08. The range represents the average value across the four markets for the years covered in the sample. 5.2 Relevant literature The main sections of the report and the appendix reference a number of sources that provide evidence for statistics and arguments made in the analysis. In addition, the study has consulted and taken into consideration other research on the subject of newspaper publishers, website traffic, and market performance. Summaries of the literature are presented below. The impact of news aggregators on internet news consumption Athey & Mobius (2012) analyse the impact of news aggregators on the quantity and composition of news in France.63 60 Results of the random effects model are available upon request. 61 For more details on panel data models please refer to Wooldridge (2010), “Econometric Analysis of Cross Section and Panel Data” 62 Cozzolino and Giarratana (2014), “Mechanisms of Value Creation in Platform Markets” 63 Athey,S. and Mobius,M. (2012), ‘The impact of news aggregators on internet news consumption: The case of localisation’ , mimeo, Harvard University. Deloitte LLP. 21