Create and explore imaginative 3D worlds with our new product Dreamfield (opens in a new tab)

Combine Bayesian model averaging with homoscedastic-subset identification

Investigate whether the Bayesian model-averaging approach for efficiently identifying subsets of homogeneous genetic variants can be adapted to identify subsets of homoscedastic variants instead.

Background

The paper proposes using heteroscedasticity tests to identify the largest subset of genetic variants for which the INSIDE assumption is plausible, thereby enabling bias correction in MR-Egger regression. Exhaustively searching all variant subsets is computationally infeasible when the number of instruments is large.

The authors contrast this approach with Bayesian model averaging, which can efficiently search for subsets of homogeneous variants and can accommodate models with multiple slopes. However, homogeneous-variant searches may fail when INSIDE holds despite all ratio estimands being different but homoscedastic. The paper therefore identifies the unresolved possibility of adapting Bayesian model averaging to search for homoscedastic, rather than homogeneous, subsets.

References

Moreover, it may be possible to combine the methods by adapting the Bayesian model averaging approach to identify subsets of homoscedastic rather than homogeneous variants, but this possibility remains to be formally investigated.

— The INSIDE assumption under all positive coding: interpretation and partial empirical assessment  (2609.37347 - Hartwig et al., 29 Sep 2026) in Section 7, Discussion, p. 24