http://www.volundarhus.com/categories/statistics/Recent content in Statistics on VolundarhusHugoen-usWed, 22 Apr 2026 00:00:00 +0000http://www.volundarhus.com/posts/smmargins/Wed, 22 Apr 2026 00:00:00 +0000http://www.volundarhus.com/posts/smmargins/<p>If you&rsquo;ve ever moved from Stata to Python and reached for <code>margins, dydx(*) at(age=(25 45 65))</code>, you know the feeling. StatsModels has <code>get_margeff</code>, but it&rsquo;s narrow: average or at-means, continuous variables, a couple of model classes. The moment you want a marginal effect <em>at a specific covariate profile</em>, or a discrete change for a factor, or a difference-in-differences on the probability scale of a logit, you&rsquo;re on your own.</p> <p>So I wrote a small module — <a href="#" ><code>marginal_effects.py</code></a> — that fills in the gap. This post is a tour of what it does, the math that makes it work, and the design decisions that kept it small.</p>