mecompare

Marginal effects and cross-model comparisons of effects in Stata

mecompare calculates marginal effects and compares them across models. It implements the framework of Mize, Doan, and Long (2019): the models are combined with suest2, so tests of cross-model differences account for the covariances of the effects across models.

Some examples of the cross-model comparisons mecompare can be used for are effects across nested models (e.g., mediation and confounding), across alternative predictors, across different outcomes, across different model types (e.g., ordinal versus nominal), and across models fit to separate groups or samples.

mecompare also calculates marginal effects within a single model and automates comparisons among them. Some examples are comparing effect sizes across predictors, tests of interaction (second differences), and the summary measures of Mize and Han (2025): the ME inequality of a nominal predictor and the Total ME across the categories of a multi-category outcome. It also calculates generalized marginal effects, in which other variables, e.g. mediators, change along with the focal variable.

More generally, mecompare can be used for most any test of marginal effects within or across models. Custom tests – any contrast, ratio, or joint test of the marginal effects – can be done with metest.

Installation

net install mecompare, from("https://tdmize.github.io/data") replace

mecompare requires Stata 16 or later and suest2, which installs the same way (net install suest2, from("https://tdmize.github.io/data") replace). metest is installed with mecompare.

Where to start

• Getting started walks through one comparison and explains the table.
• Options goes through every option with an example of each and lists the supported models; Plotting and tabulating results covers store() with coefplot and esttab.
• The Examples in the sidebar reproduce the examples of Mize, Doan, and Long (2019), Mize and Han (2025), and Mize and Han (forthcoming). Others use multilevel and panel models, including within-person effects. The Interactions pages reproduce the examples of Mize (2019) on nonlinear interaction effects.
• Examples from the help file runs the help file’s examples, and the help file is on the site.

Citation

mecompare implements the methods proposed by Mize, Doan, and Long (2019). If you use mecompare, please cite:

Mize, Trenton D., Long Doan, and J. Scott Long. 2019. “A General Framework for Comparing Predictions and Marginal Effects Across Models.” Sociological Methodology 49(1): 152–189. https://doi.org/10.1177/0081175019852763

Related commands

Command	What it does
suest2	Combines separately estimated models so that quantities from different models can be tested against each other. mecompare uses it for cross-model comparisons.
metest	Tests contrasts between rows of the tables produced by mecompare, meinequality, and totalme (or any posted margins results).
meinequality	Summarizes how unequal the marginal effects of a categorical predictor are across its categories.
totalme	Summarizes a predictor’s effect across all outcome categories of a multiple-outcome model in a single Total Marginal Effect.
suest (R)	The R version: combines separately fitted models and works with the marginaleffects package to compare predictions and effects across models.