Power priors for latent variable mediation models under small sample sizes
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Latent variable models typically require large sample sizes for acceptable efficiency and reliable convergence. Appropriate informative priors are often required for gainfully employing Bayesian analysis with small samples. Power priors are informative priors built on historical data, weighted to account for non-exchangeability with the current sample. Many extant power prior approaches are designed for manifest variable models, and are not easily adapted for latent variable models, for example, they may require integration over all model parameters. We examined two recent power prior approaches straightforward to adapt to these models, Mahalanobis weight (MW) priors based on Golchi (Use of historical individual patient data in analysis of clinical trials, 2020), and univariate priors, based on Finch (The Psychiatrist, 6, 2024, 45)'s application of Haddad et al. (Journal of Biopharmaceutical Statistics, 27, 2017, 1089) and Balcome et al. (bayesdp: Implementation of the Bayesian discount prior approach for clinical trials, 2022). We applied these approaches along with diffuse and weakly informative priors to a latent variable mediation model, under various sample sizes and non-exchangeability conditions. We compared their performances in terms of convergence, bias, efficiency, and credible interval coverage when estimating an indirect effect. Diffuse priors and the univariate approach lead to poor convergence. The weakly informative and MW approach both improved convergence and yielded reasonable estimates, but MW performed poorly under some non-exchangeable conditions. We discussed the issues with these approaches and future research directions.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.010 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it