Compare vocabulary across conditions

Calculate pairwise Jaccard similarity between the vocabularies of each condition. High overlap can help identify conditions with very similar wording. Low overlap is a reason to read the texts and ask what else may have changed.

Usage

check_lexical_overlap(texts, conditions, min_overlap = 0.1, max_overlap = 0.9)

Arguments

texts

Character vector of materials.

conditions

Condition labels.

min_overlap

Jaccard similarity below which a note is emitted (default: 0.10). Advisory only; it does not affect pass.

max_overlap

Jaccard similarity above which the check fails (default: 0.90).

Value

A list with pairwise, min_observed, max_observed, and pass.

Details

I'd treat this as a diagnostic. Vocabulary overlap also depends on text length and how many materials are in each condition. Low overlap is advisory; it doesn't fail the check.

Examples

check_lexical_overlap(
  c("The tax creates jobs and growth.", "The tax protects our children.",
    "Jobs and growth follow the tax.", "Our children deserve protection."),
  c("economic", "moral", "economic", "moral")
)
#> ✔ Distinctiveness: Jaccard overlap in [0.17, 0.17]
#> $pairwise
#> # A tibble: 1 × 3
#>   condition_1 condition_2 jaccard
#>   <chr>       <chr>         <dbl>
#> 1 economic    moral         0.167
#> 
#> $min_observed
#> [1] 0.1666667
#> 
#> $max_observed
#> [1] 0.1666667
#> 
#> $pass
#> [1] TRUE
#>