Messy data, messy conclusions: A response to Gerring

What can we learn from the analysis of a large-N observational data set (aka “messy” data)? Gerring argues that despite warnings from a number of critics, such an analysis may be deemed adequate as long as "... it allows us to update our priors, it beats the alternatives, and it presents a plausible uncertainty estimate". This seems a rigorous benchmark. Depending on our definition of plausible, even randomized trials may fail this standard when issues of treatment compliance, treatment heterogeneity, experimenter effects, or interference between units muddy the interpretation of results. Of course, observational studies may fail this standard even when such issues are not a concern. Regression results from observational studies have two additional sources of uncertainty when compared to randomized studies