Model Predictive Scoring Shows Specific BAO Observations (not SNIa) Drives w0wa Tension
Abstract
The recent analyses of DESI DR2 BAO, Planck CMB and various Supernovae datasets have shown preferences for evolving dark energy at various levels of significance, under frequentist model comparison tests. However, the same analysis done with a pure Bayesian model comparison test can lead to the opposite conclusion due to the impact of w0wa priors. In contrast to these in-sample model comparison route, we approach the problem along a third axis -- which model better predicts unseen data. We use the Expected Log Predictive Density (ELPD) metric to score the predictiveness of ΛCDM and w0waCDM models, using the leave-one-redshift-block-out cross-validation estimator. The comparison metric ΔELPD is out-of-sample, unlike Δχ2MAP and Bayes Factor, and insensitive to prior width, unlike the Bayes factor. Aggregated, we find modest preferences for w0 waCDM, primarily driven by the BAO. Decomposing the ΔELPD score by redshift we find that a single BAO block, LRG2 (z = 0.706), supplies the entire BAO preference. Interestingly, the biggest outlier point, LRG1, contributes little to the model predictive scoring because both ΛCDM and w0waCDM fail to predict the observation by equal amount. On the other hand, all SNIa datasets' ΔELPD scores are consistent with $0$, indicating that at an out-of-distribution predictive level, the w0waCDM tension is entirely driven by one BAO point, not SNIa.
Related
- The Hubble Tension: A Decade Review
- Temporal Redistribution and Flux Conservation in the Enchan Field: From Finite-Tension Phase Balance to Galactic Rotation
- Addressing the DESI DR2 phantom-crossing anomaly and enhanced H 0 tension with reconstructed scalar-tensor gravity
- Response to "Late Time Dynamical Dark Energy and the CMB-Distance Ladder Tension"
- EPRV asteroseismology of the K dwarfs GJ 570 A and GJ 783 A with ESPRESSO. Empirical mode-amplitude scaling in tension with 3D convection simulations
BibTeX
@misc{karim2026model,
title = {{Model Predictive Scoring Shows Specific BAO Observations (not SNIa) Drives $w_0w_a$ Tension}},
author = {Tanveer Karim and Joshua S. Speagle and Renée Hložek},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.04334},
eprint = {2609.04334},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.04334},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Model Predictive Scoring Shows Specific BAO Observations (not SNIa) Drives w_0w_a Tension
AU - Tanveer Karim
AU - Joshua S. Speagle
AU - Renée Hložek
PY - 2026
DA - 2026/09/03
PB - arXiv
DO - 10.48550/arxiv.2609.04334
UR - https://arxiv.org/abs/2609.04334
AB - The recent analyses of DESI DR2 BAO, Planck CMB and various Supernovae datasets have shown preferences for evolving dark energy at various levels of significance, under frequentist model comparison tests. However, the same analysis done with a pure Bayesian model comparison test can lead to the opposite conclusion due to the impact of w_0w_a priors. In contrast to these in-sample model comparison route, we approach the problem along a third axis -- which model better predicts unseen data. We use the Expected Log Predictive Density (ELPD) metric to score the predictiveness of ΛCDM and w_0w_aCDM models, using the leave-one-redshift-block-out cross-validation estimator. The comparison metric ΔELPD is out-of-sample, unlike Δχ^2_MAP and Bayes Factor, and insensitive to prior width, unlike the Bayes factor. Aggregated, we find modest preferences for w_0 w_aCDM, primarily driven by the BAO. Decomposing the ΔELPD score by redshift we find that a single BAO block, LRG2 (z = 0.706), supplies the entire BAO preference. Interestingly, the biggest outlier point, LRG1, contributes little to the model predictive scoring because both ΛCDM and w_0w_aCDM fail to predict the observation by equal amount. On the other hand, all SNIa datasets' ΔELPD scores are consistent with $0$, indicating that at an out-of-distribution predictive level, the w_0w_aCDM tension is entirely driven by one BAO point, not SNIa.
ER -