Draw a sample of materials, usually stratified by condition. I'd use this
when rating the whole pool by hand would be too expensive. Analyse the
returned ratings with human_check and compare them with
synthetic_check on the same materials.
Usage
sample_for_human_validation(
materials,
n,
stratify_by = "condition",
min_per_stratum = 1L,
text_col = "text",
seed = 42
)Arguments
- materials
A data frame of materials.
- n
Total number of materials to sample.
- stratify_by
Column name to stratify on (default:
"condition"). UseNULLfor a simple random sample.- min_per_stratum
Minimum number of materials drawn from each stratum before proportional allocation (default: 1), when the requested sample is large enough. Omitted conditions are reported. Set to 0 for purely proportional allocation.
- text_col
Name of the text column (default:
"text").- seed
Integer seed (default: 42).
Value
A tibble: the sampled rows of materials, in their original
column order, plus material_id if it was absent.
Details
When the sample is large enough, each stratum receives
min_per_stratum materials first. The remainder is allocated
proportionally using largest remainders, without exceeding the available
materials. The function reports strata that couldn't be included.
Materials with missing stratum labels are excluded and counted. The number returned is therefore capped at the number with a stratum label.
Examples
m <- data.frame(
condition = rep(c("economic", "moral"), each = 10),
text = paste("vignette", 1:20)
)
sample_for_human_validation(m, n = 6, seed = 42)
#> ✔ Sampled 6 materials from 2 of 2 stratum/strata
#> # A tibble: 6 × 3
#> condition text material_id
#> <chr> <chr> <chr>
#> 1 economic vignette 1 m0001
#> 2 economic vignette 5 m0005
#> 3 economic vignette 10 m0010
#> 4 moral vignette 12 m0012
#> 5 moral vignette 14 m0014
#> 6 moral vignette 19 m0019