where through tf are the discrete time instances at which output measurements are made. Furthermore, let y0 represent the baseline trajectory that the repeated simulations try to reproduce. Then, the standard deviation σi,j of the ith simulation of the jth configuration is defined as the standard deviation of the residuals from the baseline data,This leads to the following average standard deviations of the For the purposes of this work, it is proposed that the degradation in transparency be characterized by a statistically significant change in the average standard deviation of the residuals. On the Tradeoff between Performance and User Privacy in Information Centric Networking Formally, the null hypothesis is that all system configurations are transparent for the chosen output signal, there is no statistically significant difference between the This null hypothesis can be checked against the alternative that not all are the same using a one-way analysis of variance matrix of standard deviations On the Tradeoff between Performance and User Privacy in Information Centric Networking, with σi,j being the ith row jth column element. An ANOVA of then reveals if there is a statistically significant difference between the The outcome ofANOVAis a p-value, which is the probability of making an observation such as even more diverse than assuming that the null hypothesis is true. A high p value, typically, thus indicates that there is no statistically significant difference between the different configurations, as the differences between the can be attributed to chance. In that case, it can be concluded that the system configurations are transparent for the particular output variable considered. If the p value is low, typically if p, then the null hypothesis is rejected it is concluded that there is a statistically significant difference between the, and thus, between the configurations. In that case the conclusion is that at least one of the configurations has a statistically significant loss in transparency; however,ANOVAdoes not indicate which one. To determine which configurations are not transparent, a post hoc pairwise comparison test is necessary, and Tukey’s test is one alternative. On the Tradeoff between Performance and User Privacy in Information Centric Networking Tukey’s test considers the confidence intervals of the estimates of the column means, and concludes that two given columns, configurations, are statistically significantly different, if there is no overlap between their confidence intervals. Thus, pairwise comparisons where one of the pairs is the ideal configuration can be used to determine which configuration has a statistically significant loss in transparency. Furthermore, when a configuration is found to have a statistically significant loss in transparency, the distance between theconfidence intervals can be used as a quantification of this loss. It is important to remark that both ANOVA and Tukey’s test are parametric tests and thus, are based on some assumptions about the distributions that the data are obtained from. Specifically, the underlying assumption of both ANOVA and Tukey’s test is that the columns are independent observations coming from normal distributions with equal variances. On the Tradeoff between Performance and User Privacy in Information Centric Networking There are two ways to handle a violation of this assumption. One can either rely on the reported robustness of both tests to violations of this assumption or one can use the nonparametric counterparts of these tests, which are based on ranks and thus, do not rely on assumptions of normality The first approach is adopted in this paper; hence, formal tests of the validity of the underlying assumptions are omitted. Nevertheless, should the robustness of these methods become a concern during an analysis, Kruskal-Wallis for example, can be used instead ofANOVA, andWilcoxon rank sum test with a Bonferroni correction could replace Tukey’s test. It is worth noting, however, that such rank based tests have been found to be very conservative compared to the parametric tests employed here. On the Tradeoff between Performance and User Privacy in Information Centric Networking