Initial CRAN release.
ds.EstimateAte() orchestrates federated ATE estimation across K
independent studies using arm-specific OLS models fitted via ds.glm().
compute_variance argument (default TRUE) returns a delta-method
variance estimate and 95% confidence interval assembled from per-server
residual sums of squares and projected scatter scalars, with no
additional communication round. Set to FALSE to limit server disclosure
to the three scalars strictly needed for the point estimate (n_k,
n_k_1, tau_k).
heteroscedastic_arms argument (default FALSE) estimates arm-specific
noise variances \eqn{\sigma_1^2} and \eqn{\sigma_0^2} rather than a pooled
\eqn{\sigma^2}, adjusting the variance formula in both the delta-method
and homogeneous closed-form branches.
heterogeneous argument (default TRUE) uses the leverage-corrected
delta-method variance (valid under covariate shift and heterogeneous
treatment probabilities across sites). Set to FALSE for the simpler
homogeneous closed form.
study_effects argument (default FALSE) enables the adjusted
generalised difference estimator for settings where study membership
directly shifts the outcome. Appends K-1 integer study-dummy columns
server-side via addStudyIdDS() and adjusts degrees of freedom
accordingly.
Input validation helpers .check_datasources(), .check_args(), and
.check_string() give informative errors on bad arguments.
.check_sample_sizes() warns when either federated arm size falls
below 5*(d+1), indicating the asymptotic approximation may be unreliable.