Studies of microcredit show positive and negative treatment effects at certain quantiles of household outcome distributions. I develop new Bayesian hierarchical models to aggregate the evidence on these effects and assess their generalizability. I provide a broadly-applicable limited-information model enforcing quantile monotonicity via variable transformation. Partially discrete outcomes such as profit are aggregated using full-data mixture models. Across all outcomes I find a precise, generalizable zero effect from the 5th to 75th quantiles, and large yet heterogeneous and uncertain effects on the right tails. Households who had previously operated businesses account for the majority of the impact and the uncertainty.