Negative Binomial Regression Models.

As Fig. S2 shows, the outcome variable was heavily skewed. No substantial comments were made by new audiences during 89.35% of the organization per day observations in the dataset.

Distribution of outcome variable.

Because of the heavy skew in the outcome variable, I examined the association between the number of comments (y) an organization (i) received each day (t) and the indicators described in the main text, using a negative binomial regression model with fixed effects for time. Because fixed-effects negative binomial regression models allow for individual-specific variation in the dispersion parameter, rather than variation in the conditional mean, I also include fixed effects for each organization to facilitate unconditional estimation (19):

In addition to the indicators described in the main text of this article, the model also includes an Inverse Mills Ratio variable generated using two-stage estimation procedures to account for the modest difference in the rate of cultural betweenness among organization inside and outside the final study sample. As an additional robustness check, I ran a zero-inflated regression model, a fixed-effects Poisson regression model, and a log-linear model. Each of these alternative models produced nearly identical results.