Compute Bhat / Shat for one outcome block (joint-sigma2 form)

Marginal OLS coefficient and its joint-sigma2 standard error, suitable for the SER kernel inside SuSiE-style IBSS updates.

Replaces the previous per-(SNP, trait) marginal-residual-variance form that was breaking variational consistency. The per-SNP regressions are gone, so this is also substantially faster (one matrix multiply instead of J*p univariate fits).

Usage

cal_Bhat_Shat(
  Y,
  X,
  sigma2,
  lowc_wc = NULL,
  ind_analysis = NULL,
  v1 = NULL,
  resid_var = NULL,
  ...
)

Arguments

Y

N x J outcome matrix (one column per wavelet position or per univariate trait inside the modality).

X

N x p predictor matrix (assumed centred/scaled, but correctness does not depend on that).

sigma2

Numeric, length 1 or length J. The CURRENT IBSS residual-variance estimate for each column of Y. A scalar is recycled to length J via rep_len.

lowc_wc

Optional integer vector of column indices of Y to mask (low-count wavelet coefficients). Bhat zeroed, Shat set to 1 so they contribute nothing to BF.

ind_analysis

Optional. NULL -> use all rows. If a list, ind_analysis[[j]] is the row subset for column j of Y (per-trait missingness). If a vector, common subset for all columns.

v1

Ignored — kept for backward signature compatibility.

resid_var

Ignored — kept for backward signature compatibility. Pass sigma2 instead.

...

Swallowed.

Value

list with components Bhat (p x J) and Shat (p x J).