p. 8876, discussion of Lemma 8(b.1) — uninformed followers at mean quality exceed the average over an H- and an L-type

Let $n_{un}(\bar q)=(1-l)(1-c/\bar q)$ be the number of uninformed followers when the expected quality given the signal is $\bar q$, and $E[q]=\rho q_H+(1-\rho)q_L$. For every effort $d$ whose following cost $c=c_0+c_1(1-\tau)d$ is positive,

The left side is the expected number of uninformed followers under a separating equilibrium, the right side the number under pooling.

This is the concavity comparison the paper uses to explain why pooling yields more advertising revenue.

Formalization Note The hypothesis $c>0$ is added: with $c=0$ every uninformed consumer follows and both sides equal $1-l$, so the page's strict "higher" needs it. The "average" is the $\rho$-weighted one, as in (11).