Skip to contents

Measure O*NET Content Change Between Releases

Source: R/content_change.R
onet_content_change.Rd

Computes task-set and rating content-change metrics for each occupation between O*NET releases. This is the single source of content metrics for longitudinal work, so robustness measures cannot silently diverge. It is seam-aware: comparisons that cross the v25.1 SOC-2010 to SOC-2018 change are flagged by default so taxonomy churn is not counted as content churn. v21.0 is not a package-verified default seam; supply a custom seams row for a channel-specific v21.0 comparison.

Usage

onet_content_change(
  panel,
  scale = "IM",
  item = "task_id",
  min_importance = NULL,
  from = NULL,
  to = NULL,
  seams = NULL
)

Arguments

panel

A tibble from onet_panel() (a Task Ratings panel) or the same schema. Must contain onet_soc_code, release_version, release_date, soc_vintage, data_value, and the item key column. scale_id is required when scale is not NULL.

scale

Optional scale id used for the rating metrics and item membership, for example "IM" for Importance. Use NULL to keep every scale.

item

Column that identifies the content item within an occupation. Defaults to "task_id". Item membership drives the set metrics.

min_importance

Optional numeric floor applied to data_value on the selected scale. Items below the floor, or with missing data_value, are dropped before set membership is computed. Use it to apply an Importance floor for any channel-specific artifact a caller has external evidence for, for example near a specific release boundary.

from, to

Optional release versions selecting a single comparison. When both are supplied only that pair is returned; otherwise every adjacent release pair is compared in release-date order.

seams

Optional data frame with seam_type and seam_date columns that overrides the default seam table returned by onet_known_seams(), which currently contains only the verified v25.1 SOC-2010 to SOC-2018 taxonomy seam. Use it to supply channel-specific or source-specific seam dates, such as a v21.0 row, when a caller has external evidence that a comparison spanning that date needs seam treatment; v21.0 is not a package-verified default seam. Supply an empty table to disable date-based seams entirely. NULL keeps the default table, so Task Ratings output is unchanged. Cross-vintage SOC seams are always detected from soc_vintage regardless of this table.

Value

A tibble with one row per occupation and release pair: n_from, n_to, n_added, n_dropped, n_retained; jaccard (set similarity, the retained share of the item union); churn_rate (1 - jaccard); rating_delta_l2 (the Euclidean norm of data_value changes over retained items); cosine (cosine similarity of the data_value vectors over the item union, missing items filled with zero); and the seam flags seam, seam_type, and safely_comparable.

Details

Set metrics compare the item keys an occupation carries in each release. The filtered input must contain at most one row per occupation, item, release, and scale. Duplicate keys are rejected rather than resolved by row order. When scale = NULL and scale_id is present, item-scale combinations define membership so ratings from different scales are not joined to each other. rating_delta_l2 uses only retained items, so it measures rating drift among tasks that persist; cosine uses the union with zero fill, so it responds to both membership turnover and rating change. Seam-crossing pairs still receive metrics but are marked safely_comparable = FALSE; include them only in a clearly labeled seam-inclusive row. Occupations are matched on onet_soc_code, so cross-taxonomy pairs generally share few codes and should be bridged with onet_crosswalk_bridge() before interpretation.

Examples

panel <- tibble::tibble(
  release_version = c(rep("22.1", 3), rep("23.1", 3)),
  release_date = c(rep(as.Date("2017-10-01"), 3), rep(as.Date("2018-11-01"), 3)),
  soc_vintage = "2010",
  onet_soc_code = "15-1132.00",
  task_id = c("1", "2", "3", "1", "2", "4"),
  scale_id = "IM",
  data_value = c(4.0, 3.5, 2.0, 4.5, 3.5, 3.0)
)
onet_content_change(panel)
#> # A tibble: 1 × 21
#>   onet_soc_code soc_code title from_release to_release from_release_date
#>   <chr>         <chr>    <chr> <chr>        <chr>      <date>           
#> 1 15-1132.00    15-1132  NA    22.1         23.1       2017-10-01       
#> # ℹ 15 more variables: to_release_date <date>, soc_vintage_from <fct>,
#> #   soc_vintage_to <fct>, n_from <int>, n_to <int>, n_added <int>,
#> #   n_dropped <int>, n_retained <int>, jaccard <dbl>, churn_rate <dbl>,
#> #   rating_delta_l2 <dbl>, cosine <dbl>, seam <lgl>, seam_type <chr>,
#> #   safely_comparable <lgl>