Unlike a methods textbook that spends 300 pages on the properties of MLE, Bellemare’s PDF tells you exactly what to click, what to write in your Stata/R do-file, and what to say in your dissertation defense when a committee member asks, “Did you check for outliers?”
While you should read the original PDF for the full nuance, here is the skeleton of Bellemare’s framework:
One of the most influential sections of the guide deals with Pre-Analysis Plans. While common in development economics (due to RCTs), Bellemare argues that PAPs are useful for any empirical project. The PDF explains how to write a PAP that specifies your hypothesis, your empirical strategy, and your inclusion/exclusion criteria before you look at the data. This prevents the cardinal sin: fishing for statistically significant results.
Before diving into the document itself, it is crucial to understand the author. Marc F. Bellemare is a Distinguished McKnight University Professor in the Department of Applied Economics at the University of Minnesota. He holds appointments in the Department of Economics and the Humphrey School of Public Affairs.
Bellemare is not an armchair theorist. He is an applied economist who has published extensively on agricultural economics, food security, political economy, and the economics of new technologies (e.g., UAVs in agriculture). He is also famous for his rigorous, no-nonsense approach to causal inference.
More importantly for this discussion, Bellemare is one of the most transparent and generous economists on the internet. He regularly posts drafts, replication files, and advice on his personal website. His blog is a goldmine for grad students. The “Doing Economics” document originated from a guest lecture he gave in a PhD field course. Because it filled a massive gap in formal training, he made it available online as a PDF – and the field has never been the same.
In the "Doing Economics" PDF, Bellemare is ruthless about identification. He famously distills the issue into a simple question: Does your empirical strategy recover the parameter of interest? If you use an instrumental variable (IV), does it actually satisfy the exclusion restriction? If you use difference-in-differences (DiD), is the parallel trends assumption plausible? The guide provides plain-English ways to defend your strategy.