Which relative effects of a component network are uniquely estimable?

The additive component model identifies a relative effect theta_i - theta_j = (C_i - C_j)' beta only when the contrast vector C_i - C_j lies in the row space of the design matrix X = B C (Wigle et al. 2026). Full column rank of X (rank equal to the number of components) is sufficient for every contrast to be estimable, but it is not necessary: a disconnected, rank-deficient component network can still identify many cross-sub-network treatment contrasts.

Usage

estimable_effects(object, reference = NULL, ...)

Arguments

object: A cpaic_network(), cpaic_connectivity(), cpaic_bridge or cpaic_mlnmr object.
reference: Reference treatment. Defaults to the network reference.
...: Unused.

Value

A data frame with one row per treatment, giving the treatment, the comparator (the reference), and estimable (logical).

Details

Checking this matters because both engines otherwise return a finite-looking answer for a contrast that carries no information: the frequentist weighted least squares through the Moore-Penrose pseudoinverse, and the Bayesian model through the prior.

References

Wigle A, Beliveau A, Nikolakopoulou A, Lin L (2026). Creating Treatment and Component Hierarchies in Component Network Meta-Analysis.

Examples

net <- cpaic_network(cpaic_bin_agd, sm = "OR", inactive = "Placebo")
estimable_effects(net)
#>   treatment comparator estimable
#> 1         A    Placebo      TRUE
#> 2       A+B    Placebo      TRUE
#> 3     A+B+C    Placebo      TRUE
#> 4     A+B+D    Placebo      TRUE
#> 5         B    Placebo      TRUE