Online Appendix: Ideological Segregation Online and Offline

In table 2 we present standard errors for our main estimates. For estimates that come from MRI, the GSS or the CNES, we compute standard errors through a nonparametric bootstrap over samples of respondents. For estimates that come from comScore aggregate data, that is not an option. Instead we perform a parametric bootstrap in which we model the share conservative of each website as an independent binomial proportion with sample size determined by the size of the relevant comScore panel. (In the model appendix to the paper, we show that standard errors from a model that relaxes the independence assumption are on the same order of magnitude as those we present in table 2.)