Home | Aim and scope | Call for Papers | Registration | Important Dates | Programme | Workshop | Speakers | Instructions for Speakers | Travel and AccommodationSponsors | Committees | Contacts

Workshop on Multilevel regression and poststratification (MRP), 21 August 2023, 12 to 4 pm

The half-day workshop on Multilevel regression and poststratification (MRP) will take place as part of the conference on Monday 21 August. Participation (onsite and online) is free of charge.

Speaker: Philipp Christian Broniecki, University of Oslo, Norway

autoMrP is a tool for estimating public opinion in subnational units, such as states, from nationally representative surveys. It combines multilevel regression with post-stratification (MrP) and machine learning. MrP has become the gold standard for small area estimation. Researchers who want to apply MrP are faced with three problems: how to select features, how to specify the functional form, and how to regularize the model parameters. autoMrP, an R package available on CRAN (, addresses these problems in a systematic way using machine learning techniques. The method is described in detail in Broniecki, P., Leemann, L., & Wüest, R. (2022). Improved Multilevel Regression with Poststratification through Machine Learning (autoMrP). The Journal of Politics, 84(1), 597-601.

Time: Monday 21 August 2023, 12 to 4 pm (Time zone: UTC+3)

Place: Soc&kom, Festsal, University of Helsinki and Zoom

Registration form (NOTE. If you have already registered for the BaNoCoSS conference, you do not need to register again.): Registration is required for both in-person and online (virtual) participants. Link to the online platform used to share materials etc. will be communicated and opened for registered participants before the event.

Bring your own laptop. Welcome!

Dropbox link:

Outline autoMrP hands-on course

Have R installed (preferably 4.x). We install packages from CRAN. In section 4 of the course, we apply autoMrP. Bring your questions and/or data if you have thought about working with autoMrP in your project.

1) MrP, regularization, and the problem autoMrP attempts to solve (12 – 12.50 pm)

  • Review the MrP approach, the random intercepts model with post-stratification
  • How partial pooling regularizes
  • The problem that autoMrP attempts to solve
  • Practical: In R, estimate an MrP model using the autoMrP package

 10 minutes break

 2) Machine learning for prediction tasks (1 – 1.50 pm)

  • Review central concepts: bias-variance trade-off, cross-validation, loss-functions
  • Review best subset, k nearest neighbors, the lasso, support vector machine, gradient boosting
  • Review ensembles
  • autoMrP recipe
  • Practical: Use autoMrP

 10 minutes break

 3) autoMrP in practice I (2 – 2.50 pm)

  • how to construct the census data
  • how to input custom folds
  • uncertainty

 10 minutes break

 4) autoMrP in practice II (3 – 4 pm)

  • complex sampling designs (weighting, clustering)
  • lopsided dependent variables
  • when not to use autoMrP and how to tell
  • alternatives to autoMrP
  • questions/solving problems with your data


Last update 21 August 2023

  • No labels