When you design experiments, you need to know how many participants it takes to get informative results. But what makes results informative? Simply put: A precise effect estimate, meaning an estimate that is unbiased and has a narrow confidence interval. Your randomized design and careful measurement should ensure that your effect estimate is unbiased. But how can you be confident that your estimate’s confidence interval is narrow ’enough'?

Continue reading

When preparing the analysis for our recent preregistered field experiment on the effect of carbon foodprint labels, we decided to implement an (interactive) specification curve analysis to present insights on how large our researcher degrees of freedom were, even after preregistration. This blog post introduces the idea and links you to the GitHub repo, where you can reproduce and explore our analyis relatively quickly using GitHub Codepaces.

Continue reading

Building on a current working papers of David Veenman and myself, and using ‘fancy’ animations, we discuss the issues related to non-random outliers in empirical archival research work and whether robust regression methods can be viewed as a pancea (spoiler: they can’t). Building on these insights we suggest a work-flow for archival work that helps us to take outlier treatment to the next level.

Continue reading

Designing empirical research to be easily reproducible is no easy task. The new ’treat’ repository, developed by the Open Science Data Center of the TRR 266 Accounting for Transparency, accompanied by a series of short videos, leads the way, step by step.

Continue reading

I had it on my long-term to-do list: Understanding how the standard errors that the Stata command ‘reghdfe’ generate differ from the standard errors that various R package for panel fixed effect models generate. Here is what I learned.

Continue reading

I recently included the new Our World in Data data on Covid-19 vaccination progress around the world in the {tidycovid19} package. What was meant to be a short info post for package users turned into a mini case on “outliers”. See for yourself

Continue reading

It started as a Christmas present and ended up to be a fun case study on the productive boost of the open source community. Use R and shiny to hack the LEGO Art Beatles set to become a Stones set!

Continue reading