Symptom selection across the plateau, against chance

Reports how often each PCL-5 item appears among the near-optimal symptom combinations, alongside the frequency expected by chance alone. Raw selection counts cannot be read on their own: every combination contains the same number of items, so the counts are zero-sum and an item can only gain at another's expense. The chance baseline turns "item 6 appears in 55 plateau combinations" into a statement about whether item 6 is actually being preferred.

Usage

symptom_selection(
  fit,
  delta = 1,
  stability = NULL,
  clusters = NULL,
  symptom_labels = NULL
)

Arguments

fit

A fitted exhaustive search from score_all_combinations.

delta

Numeric. Plateau width in percentage points of balanced accuracy (default 1). A single value.

stability

Optional bootstrap_stability result for the same search, used for the stability weighted columns.

clusters

Named list of item indices used only to label items by cluster in the output. Defaults to the clusters of fit, or to the standard DSM-5-TR PCL-5 clusters when fit was non-hierarchical.

symptom_labels

Optional character vector of length 20 giving readable item names.

Value

A data.frame with one row per item:

• symptom, label, cluster: item identity.

• baseline: proportion of this rule form's candidate space containing the item.

• plateau_count, plateau_n, plateau_freq: how often the item appears among the plateau combinations.

• enrichment: plateau_freq / baseline. Values above 1 mean the item is selected more often than chance would produce.

• excess_pp: plateau_freq - baseline in percentage points.

• pi_freq, pi_enrichment: the stability weighted equivalents, or NA when stability is not supplied.

Details

The baseline is the proportion of the rule form's own candidate space that contains each item, computed by enumeration rather than assumed. Under the non-hierarchical rule it is the same for every item (n_symptoms / 20, so 30 hierarchical rule it varies substantially by cluster, because a combination must draw at least one item from each: with the default PCL-5 clusters, each of the two avoidance items appears in 55.8 each of the seven cognition-and-mood items appears in only 25.2 hierarchical selection frequencies against a flat 30 avoidance items look preferred when they are merely unavoidable.

Supplying a bootstrap_stability result adds a stability weighted version of the same quantity, in which each combination counts in proportion to how often it stayed on the plateau across bootstrap replicates rather than counting once if it happened to make the plateau in the original sample.

See also

compute_plateau, bootstrap_stability, plot_symptom_selection.

Examples

# \donttest{
ptsd_data <- rename_ptsd_columns(simulated_ptsd[1:120, ],
                                 id_col = c("patient_id", "age", "sex"))
fit <- score_all_combinations(ptsd_data, n_symptoms = 4, n_required = 3,
                              show_progress = FALSE)

selection <- symptom_selection(fit, delta = 1)
head(selection[order(-selection$enrichment), ])
#>    symptom label cluster baseline plateau_count plateau_n plateau_freq
#> 6        6  <NA>       C      0.2             8         8        1.000
#> 7        7  <NA>       C      0.2             8         8        1.000
#> 11      11  <NA>       D      0.2             5         8        0.625
#> 12      12  <NA>       D      0.2             4         8        0.500
#> 4        4  <NA>       B      0.2             2         8        0.250
#> 17      17  <NA>       E      0.2             2         8        0.250
#>    enrichment excess_pp pi_freq pi_enrichment
#> 6       5.000      80.0      NA            NA
#> 7       5.000      80.0      NA            NA
#> 11      3.125      42.5      NA            NA
#> 12      2.500      30.0      NA            NA
#> 4       1.250       5.0      NA            NA
#> 17      1.250       5.0      NA            NA
# }