Skip to contents

Simulated scores of 3,000 Grade 8 students (1,500 in 2025 and 1,500 in 2026, a quarter of them English language learners, ELL) on a science test with 16 multiple-choice items and 4 constructed-response items, as scored operationally: The constructed responses (CR1 to CR4, 0 to 3 points) are scored by an automated scoring engine. The human scores of the validity sample are in aiscored_cr.

Usage

aiscored

Format

A data frame with 3,000 rows (one per student) and 23 columns: student, year (2025 or 2026), group (factor, Fluent or ELL), MC01 to MC16 (0/1) and CR1 to CR4 (machine scores, 0 to 3).

Source

Simulated; the script is data-raw/aiscored.R in the package's GitHub repository.

Details

The dataset is for learning to audit an AI component with ai_audit(). Some problems were planted on purpose; lab_truth("aiscored") lists them and the true scores.

See also

Examples

table(aiscored$year, aiscored$group)
#>       
#>        Fluent  ELL
#>   2025   1148  352
#>   2026   1129  371
colMeans(aiscored[, c("CR1", "CR2", "CR3", "CR4")])
#>      CR1      CR2      CR3      CR4 
#> 1.441333 1.261667 1.751333 1.402000