James Bland

Thanks to the @ecscienceassoc.bsky.social team for an amazing N American meeting. I am glad that there is room for me in this space!

7 reasons to use Bayesian inference!
statmodeling.stat.columbia.edu/2025/10/11/7...

❗️Our next workshop will be on Oct 16th, 6 pm CEST titled Structural Bayesian Techniques for Experimental and Behavioral Economics in R& Stan by @jamesbland.bsky.social
Register or sponsor a student by donating to support Ukraine!
Details: bit.ly/3wBeY4S
Please share!
#AcademicSky #EconSky #RStats

One week until this online workshop on my book!

a cartoon of homer simpson with the words which was the style at the time ALT: a cartoon of homer simpson with the words which was the style at the time

Proposed new language for "which is how it is done in the literature"

Is this something you could feasibly do? Yes! My computer designed these experiments in less than a day. It is cheap compared to paying for a less good experiment.

So should I use this technique for designing experiments? I would say yes with a but: Yes, but the algorithm only understands the econometrics. It doesn't really understand humans the same way experimenters do. So in the end it should just be used as a suggestion.

Admittedly I don't gain any intuition along the way, but my computer provides suggestions for designs that I couldn't have thought of myself.

The take-away? The best experiment design depends on the research question. In some sense: duh. But in another sense going into this I had absolutely no intuition as to how these should differ.

The designs are really different to each other, and also look (to me at least) very different to existing designs by humans.

Each of these experiments consists of 80 pairwise choices over lotteries with 3 prizes. There are 320 design variables!

- One to estimate all of the parameters as precisely as possible,

- another to estimate just the probability-weighting parameters as precisely as possible, and

- another to estimate a certainty equivalent.

I demonstrate this approach by designing three experiments to estimate a rank-dependent utility model. ...

Once we've written down (and probably also approximated) our utility function, we then need to maximize it. This is computationally non-trivial: the choice set for experimenters is very large! I tackle this with an exchange algorithm.

Furthermore, in structural estimation we are often interested in transformations of the parameters. What if we wanted to estimate one of these transformations well? My approach also lets you do that!

For example, what if your model had 4 parameters, but you only really wanted to estimate and comment on 1 of them? Surely sacrificing some of the precision of the remaining 3 in favor of the one parameter you care about is a good idea.

Yes. It is. My approach lets you do that!

Firstly, how do we even come up with the utility function V(d)? There are some out-of-the-box suggestions out there like D-optimal design that maximize the determinant of the information matrix. This maximizes the precision of your estimator, but I wanted to be more specific than this. ...

I approach this problem as a utility-maximization problem for the experimenter. Let V(d) be the experimenter's ex-ante experiment design d.

Conceptually this is a simple, standard economic problem: choose d (subject to constraints) to maximize V

But there are many practical issues with it ...

Big picture idea:

Q: What should you do if you plan on estimating a structural model from a new economic experiment?

A: it depends on what question you're trying to answer!

This paper: how to design your experiment to best answer your research question (if you're using a structural model)

New version of a working paper:

"Optimizing experiment design for estimating parametric models in economic experiments"

github.com/JamesBlandEc...

#EconSky

A thread ...

Have you got a particular application in mind?

The applications will be economics-related. It is based on material from my book:

jamesblandecon.github.io/StructuralBa...

That said, if you are wanting to estimate structural models from data using Bayesian techniques it will hopefully provide a good primer

Another new blog post!

This time, I resurrect some data collected with my kids during COVID and give it the structural Bayes treatment

#EconSky #RStats #Stan

jamesblandecon.github.io/posts/2025-0...

Working on another blog post. It's another angle accuracy model, but this time with Nerf darts and data I collected with the kids during COVID lockdown.

#RStats #EconSky

I care so little about this kind of football. I had to look up the kickers in my top 10 ranking on Wikipedia to make sure they were plausible 🤣🤣. But there's very good data on it, and I felt like crunching some numbers today.

A new blog post!

NFL field goal attempts: Geometry and Stan

This is an extension of some work I did a few years ago. Now I am using newer data and a richer statistical model.

#RStats

jamesblandecon.github.io/posts/2025-0...

In some non-econ news, I saw The Darkness in Detroit last night. Much fun was had!

Link to book here: jamesblandecon.github.io/StructuralBa...

Dariia Mykhailyshyna - Workshops for Ukraine Feedback on the past workshops (if you want to learn how to make wordclouds, check out Text Data Analysis workshop below)

On October 16th I will be presenting material from my book in the Workshops for Ukraine online workshop series. Details here 👇👇

sites.google.com/view/dariia-...

#EconSky #RStats #TeachBE