Normalised inverse-observation weights for support assessment

Construct the averaging measure at one exposure position from the conditional probabilities of its preceding observation transitions.

Usage

margot_lmtp_support_weights(probability, reached, row_id)

Arguments

probability

Numeric matrix of conditional observation probabilities, with named rows equal to row_id and one named column per required preceding transition. Values outside reached are ignored.

reached

Named logical vector identifying records observed at the current exposure.

row_id

Unique nonempty character identifiers in input order.

Value

A list with state, normalised weight, cumulative log_inverse_observation, observation_effective_n, and reached count n. Unreached vector entries are NA. Invalid probabilities, identities, an empty risk set or numerical underflow return state unassessable with a reason.

Details

Products are accumulated in log space and rescaled by one common constant. No probability floor or weight cap is applied. All reached records enter normalisation, regardless of their exposure ratios.

Examples

ids <- c("a", "b")
p <- matrix(c(1, .25), dimnames = list(ids, "observation"))
margot_lmtp_support_weights(p, setNames(c(TRUE, TRUE), ids), ids)
#> $state
#> [1] "assessable"
#> 
#> $weight
#>   a   b 
#> 0.2 0.8 
#> 
#> $log_inverse_observation
#>        a        b 
#> 0.000000 1.386294 
#> 
#> $observation_effective_n
#> [1] 1.470588
#> 
#> $n
#> [1] 2
#>