Original title: A Cumulative Ordered Spike-and-Slab Prior for Adaptive Dimension Selection in Joint Latent Space Models
Bin Lv, Yincai Tang, Siliang Zhang · September 2026
Psychologists often study complex networks, like how people connect, alongside extra traits for each person. A joint latent space model maps this data into a hidden space where nearby points show strong links. The problem is knowing how many dimensions—or hidden directions—this map needs. Usually, scientists must guess the count beforehand or run slow extra tests. To fix this, researchers created a Bayesian method using a cumulative ordered spike-and-slab prior. This mathematical rule systematically tests each possible dimension in order, turning off unneeded ones. It infers the true dimension automatically alongside all model parameters. In tests, it correctly finds the right number of dimensions, accurately estimates parameters, and runs efficiently without guessing.
Imagine you are trying to find the best number of shelves for a bookcase. You start with too many shelves, but you don't know which ones to remove. Normally you would have to guess or try every combination. Here, the bookcase itself tells you which shelves are unnecessary as you build it, by automatically collapsing the ones that hold no books. In this paper, the latent dimensions are like shelves, and the spike-and-slab prior prunes away dimensions that do not help explain the data.
Psychometricians analyzing network data with many node attributes no longer need to guess the number of latent dimensions or run costly post-hoc tests. This method automatically finds the right dimension while fitting the model, saving time and improving accuracy in understanding relationships like social ties or test item responses.
The abstract does not discuss limitations, but the method relies on MCMC sampling, which may be computationally intensive for very large networks. It also assumes the joint latent space framework, which may not suit all data types.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
A Cumulative Ordered Spike-and-Slab Prior for Adaptive Dimension Selection in Joint Latent Space Models — Bin Lv, Yincai Tang, Siliang Zhang — Psychometrika — doi:10.1017/psy.2026.10134
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.