In many scenarios, data are structured in groups or clusters, e.g. pupils within classes (within schools), survey respondents within countries or, for longitudinal surveys, survey answers per subject. Simply ignoring this structure will likely lead to spuriously low standard errors, i.e. a misleadingly precise estimate of our coefficients. This in turn leads to overly-narrow confidence intervals, overly-low p-values and possibly wrong conclusions.

*Clustered standard errors* are a common way to deal with this problem. Unlike Stata, R doesn’t have built-in functionality to estimate clustered standard errors. There are several packages though that add this functionality and this article will introduce three of them, explaining how they can be used and what their advantages and disadvantages are. Before that, I will outline the theory behind (clustered) standard errors for linear regression. The last section is used for a performance comparison between the three presented packages. If you’re already familiar with the concept of clustered standard errors, you may skip to the hands-on part right away.