Paid MRP project (video-game industry pay report, 5k salary obs)

Jobs
1

Hi all,

I’m Julie, CEO of Hushcrasher, a small data science company covering the video-game industry (we publish our analyses in a newsletter). We bring academic methods to industry questions (my cofounder Antoine Mayerowitz holds an economics PhD from the Collège de France under Philippe Aghion).

We’re looking for someone to own the estimation pipeline for a report on games-industry pay, as a paid, well-scoped consulting engagement.

The problem: We’ve pooled ~5,000 salary observations from several surveys spanning 2019-2025. The instruments are heterogeneous (differing question wording and response formats). Rather than reporting raw sample means, we want to reweight to the real population of games workers (country × job family × gender x etc). We want to do it properly (MRP / poststratification, partial pooling, honest uncertainty, sensible handling of sparse cells). Open, of course, to hearing where the method should push back on our framing.

The engagement covers the full pipeline: harmonization → multilevel model + poststratification → the statistical and editorial calls (which subgroups we can speak to confidently, what to aggregate vs. break out) → visualization.

Why we do this: The video game industry has been through repeated layoffs since 2023. Power sits with employers. This is a free, public report meant to reduce the pay information asymmetry and give workers a bit of leverage back.

Logistics: Remote, paid per day. Publication targeted for late September / early October, so we’d like to start soon. Open to other arrangements too, we already collaborate with academics (Harvard, Mannheim) on a data-for-research basis.

If the problem or the mission resonates, reply here or email me at julie@hushcrasher.com. Happy to share more detail on the data.

2

Thanks for posing, @JulieBelzanne, and welcome to the Stan forums.

This is something that many of our users could handle. If the data’s clean and the covariates are well described, this shouldn’t take long at all. Also, feel free to ask for advice here—we talk about these topics all the time.

P.S. The multilevel regression (MR in MRP) implements a partial pooling approach to sparse cells and the postsratification (P) handles missing at random observations in unbalanced stratified designs that don’t match the population. Bayesian inference should handle the uncertainty quantification (I don’t know what “honest” means in this context apart from calibration).

The main thing that’s going to be hard to control here is how the sample was constructed. If there’s some bias in the collection (e.g., people with higher pay respond less, people happy in their job respond less, people who just changed jobs respond more, etc.), then the trick’s going to be getting some handle on that. When we were working on estimating Covid prevalence, for example, we had lots of granular data on PCR test results from the UK government, but had no good way to control for the fact that people with symptoms were more likely to get tested.

P.S. Shira’s been doing a great series of posts on survey sampling covering all these issues and providing references on Gelman’s blog. The trick’s going to be not turning this into a Ph.D. thesis!

P.P.S. One of our former postdocs and a Stan developer, @bbbales2, works at Riot Games.