Skip to content

Conversation

@jeliason
Copy link
Owner

No description provided.

Implements conditional variance structure to address high uncertainty
in spatial interaction curves (SICs) when patients have only one image.

Changes:
- Preprocessing detects single-image patients and passes flag to Stan
- Stan models collapse image-level variance for single-image patients
- Multi-image patients retain full 3-level hierarchical structure

This resolves identifiability issues where image and patient levels
were confounded, reducing unnecessary uncertainty in estimates.
When num_pt_groups == 1, the model cannot estimate between-group
variance since there's no replication at the group level. This commit
modifies the hierarchy to:

- Estimate beta_global as fixed population mean with direct prior
- Skip estimation of sigma_beta_global (undefined with one group)
- Patients still vary around the group mean via sigma_beta_indiv

Changes apply to both centered (SHADE.stan) and non-centered
(SHADE_ncp.stan) parameterizations.

Three cases now handled:
- num_pt_groups > 1: Full hierarchy with between-group variance
- num_pt_groups == 1: Fixed group mean, patient variation only
- num_pt_groups == 0: Two-level hierarchy (patient + image)
@jeliason jeliason merged commit d956614 into main Oct 27, 2025
5 checks passed
@jeliason jeliason deleted the adaptive-hierarchy-single-image branch October 27, 2025 22:34
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants