Operational work needs decisions you can defend and reports that non-statisticians can read. mirtverse wraps mirt’s estimation in a documented workflow: Each step checks the results and raises flags that a person reviews before the next step.
Your path in three steps
Read the case study first. It follows a fictional certification board through eight decisions and ends with a technical report you can download.
1. Walk through the assessment pipeline.
mv_pipeline() takes a test through 13 stages, from the
blueprint to bank monitoring. The pipeline article runs a full
example:
library(mirtverse)
# certexam is a built-in certification exam with a scoring key
p <- mv_pipeline("Certification exam, form A")
p <- add_responses(p, certexam, items = 3:34, key = certexam_key)
p <- calibrate(p) # fits the model and flags miskeys, misfit and DIF
flags(p) # what needs a decision before you move on2. Check fairness and precision. DIF in one call, item health cards for review meetings, and information and conditional precision at the cut score:
mod <- p$model # an ordinary mirt model
item_card(mod, "Q14") # one-page review of a flagged item
res <- dif(scored, group = gender) # your scored 0/1 data and a grouping variable
gg_dif(res)3. Report to people who do not use R. A technical report from the pipeline, a static HTML report, or an interactive dashboard that runs in any browser:
pipeline_report(p, file = "technical-report.html")
export_dashboard(mod, "results.html", format = "html")Open the live example dashboard to see what your stakeholders would get.
Why you can trust the numbers
- mirtverse never changes how models are estimated; every result comes
from mirt, and every function accepts an ordinary
mirt()model. - The package has an automated test suite and passes
R CMD checkon every change (check status). - Every flag rule is listed by
mv_rules()and can be changed, and every decision you acknowledge is recorded in the report. - The workbench and
mv_ask()show the R code they run, so analyses can be reviewed and rerun.
Related
-
Response times and rapid guessing
with
fit_rt(). - Linking and equating and item banks and adaptive testing in the book.
- Auditing AI components if your program uses automated scoring.