Skip to contents

Changelog

Source: NEWS.md

causalgenerics 0.0.0.9000

  • Added new_causal_wts(), the low-level constructor for the abstract causal_wts weight class that concrete weight classes in the ecosystem are built on.

  • Added causal_wts methods for is_causal_wt(), estimand(), and estimand<-(), so a concrete class inherits all three.

  • Added causal_wts methods for the read-only operations that name no metadata field and so give the answer the underlying double would: vec_math(), the Summary group generic, min(), max(), range(), median(), quantile(), summary(), anyDuplicated(), diff(), [, and the six comparison operators. Most of them unwrap the weights and hand the work to the base or vctrs implementation; [ instead delegates to the next method, except when the index is a matrix or an array. Either way, the concrete class’s own vec_restore() stays in charge of the metadata wherever the result is still a weight vector.

  • Added ess.default(), which computes the Kish effective sample size for any numeric vector. Weight vectors are numeric underneath, so a concrete weight class inherits a working ess() without writing a method. The default rejects input that is not numeric, including NULL, and returns NaN for numeric input with no information to summarize, such as a zero-length vector or weights that are all zero.

  • Added causal_wts_ptype2(), the coercion rule that a concrete weight class calls from its own vec_ptype2() method for two vectors of that class: they combine only when their estimands are identical, and otherwise the common type is a plain double. The condition signalled on the downgrade path stays with the caller.

  • Added new_ipw(), the low-level constructor for the object every ipw() method returns, together with print() and as.data.frame() methods for the ipw class. The field names and their order are now a cross-package contract, so an IPW estimate reads the same way whichever package produced it and a package supplying an ipw() method inherits both methods rather than writing its own. as.data.frame() gains an exponentiate argument that moves the log(rr) and log(or) rows to their natural scale, exponentiating the point estimate and the confidence limits only. Standard errors, z statistics, and p-values stay on the log scale, where the inference is done.

causalgenerics 0.0.0.9000