If (\lambda = 0.1), threshold (p=0.2). If estimated (p < 0.2), they stop early. Families observe historical stops and national ratio changes. Using Bayesian learning, after several days they form a posterior on (\lambda). This influences future stopping.
where (k > 0) is a sensitivity parameter (here, (k=2)). the hardest interview 2
If (\Delta U < 0), they stop even if formal stopping rule not met (early stop). [ U_\texttotal = \sum_\textfamilies \left( \fracb_fg_f - \lambda \cdot t_f \right) ] If (\lambda = 0
[ R_n = \fracB_nG_n,\quad B_n = B_n-1 + X_n,\ G_n = G_n-1 + (1-X_n) ] where (X_n \sim \textBernoulli(p_n)). If (\lambda = 0.1)
[ p_n = \frac11 + e^-k \cdot (R_n-1 - 1) ]