Assemble per-respondent comparison matrices

Builds one square judgement matrix per respondent from the pair columns of a "pairwise_comparison" item. A "saaty" item stores one signed integer per unordered pair, so the reciprocal half of each matrix is reconstructed here rather than stored (a collected value of +5 becomes m[a, b] = 5 and m[b, a] = 1/5, and the diagonal is 1). An "influence" item stores one unsigned integer per ordered pair, fills the directed cell only, and has a zero diagonal, because the influence of a on b says nothing about the influence of b on a.

Usage

sframe_assemble_pairwise(data, instrument, item_id)

Arguments

data

A data frame of responses, as produced by read_responses().

instrument

An sframe instrument declaring item_id.

item_id

Character. The id of a "pairwise_comparison" item.

Value

A list with matrices (a list of square numeric matrices with dimnames), n_respondents, n_dropped, dropped (a data frame of row number and reason), items, and scale.

Details

Respondents who left any pair blank, or whose answer falls outside the declared scale, are dropped whole and counted. Completing partial matrices (the Harker approach) is deliberately out of scope.

See also

sframe_aggregate_judgements(), sframe_collected_weights()

Examples

crits <- c("price", "speed")
study <- sf_instrument(
  title = "Demo", components = list(
    sf_item("pairs", "Compare the criteria", type = "pairwise_comparison",
            comparison_items = crits)
  )
)
responses <- data.frame(pairs__price__vs__speed = c(3, -5))
sframe_assemble_pairwise(responses, study, "pairs")$matrices[[1]]
#>           price speed
#> price 1.0000000     3
#> speed 0.3333333     1