A Practitioner’s Guide to Cluster-Robust Inference

By Colin Cameron, Douglas L. Miller. The Journal of human resources, 2015

4,501 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/c5961a9a3f18259c45e684bc4a44f84943b970bf

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Statistics and Probability

Citations in the corpus

The Impact of Cash and Food Transfers: Evidence from a Randomized Intervention in Niger (2018, positive): “As Cameron and Miller (2015) write, “There is no clear-cut definition of ‘few’; depending on the situation ‘few’ may range from less than 20 to less than 50 clusters . . .” At 52 clusters, arguably we are at the boundary of “few” clusters, but given that we have unbalanced clusters, we assessed the…”
Cluster-Robust Bootstrap Inference in Quantile Regression Models (2014, background): “Cluster-robust inference in linear regression has a long history in economics; see Cameron and Miller (2015) for a recent survey.”
How Media Coverage of Corporate Social Irresponsibility Increases Financial Risk (2017, neutral): “In deciding how to assign clusters, we followed the advice of Cameron and Miller (2013) to use clustering units that are larger than the primary sampling unit, but not so large that there would be a concern about having too few clusters.”
Small-Sample Methods for Cluster-Robust Variance Estimation and Hypothesis Testing in Fixed Effects Models (2016, methodology): “Following Cameron and Miller (2015), we refer to this estimator as CR0.”
Inference in Differences-in-Differences with Few Treated Groups and Heteroskedasticity (2019, methodology): “there is evidence that it should provide a better finite sample approximation (see Davison and Hinkley (1997), Cameron et al. (2008), and Cameron and Miller (2015)).”
On the Unnecessary Ubiquity of Hierarchical Linear Modeling (2017, background): “…results in estimates of Var ̂ being too small because covariances between observations are constrained to zero, meaning that the standard errors are underestimated because terms from the most general formula should not cancel, and, consequently, Type I errors are inflated (Cameron & Miller, 2015).”