EvalIA
Selected work · Leer en español
Professional tools · Psychometric analysis
EvalIA
A psychometric analysis tool that keeps the decisions behind each result.
The question
EvalIA connects data preparation, reliability, factor analysis and six IRT models with reports and reproducible R code.
A key decision
The tool checks results against independent equations and rejects inadmissible solutions even when they converge. Changes to data or report configuration invalidate earlier outputs to avoid delivering an outdated document.
My role
I define the psychometric requirements and workflow, and direct implementation and review with AI agents.
What was checked, and how
The September 11, 2026 record separates independent equations, public references and integration checks. This extract preserves the stated tolerances; it is not a new execution.
| Output | Reference | Documented scope |
|---|---|---|
| IRT probabilities | Equations outside mirt | Rasch, 2PL, 3PL, GRM, PCM and RSM; 8 items and 19 θ values per model; tolerance 1e−8 |
| IRT information | Numerical derivatives | Six models; tolerance 1e−6 |
| EAP and posterior error | Integration over 8,001 points | Three patterns per model; tolerance 0.003 |
| Reports | Data or configuration changes | Earlier downloads invalidated; workflows with synthetic data |
One design decision: convergence is insufficient to accept a CFA solution with a negative residual variance. The tool rejects that solution. It also avoids presenting a report as current after its configuration changes.
Full validation record (Spanish, September 11, 2026) · Workflow and state checks (Spanish) · Provenance
Validation documents specific calculations and contracts. It does not certify instruments, populations or every possible case. The record retains two historical omissions in the main suite and does not replace evaluation with external professionals.
Evidence status
- Checked / documented
- The September 11 record documents probabilities for six IRT models (tolerance 1e−8), information (1e−6) and EAP (0.003), plus reliability and factor analysis. Phase 4 records 15 UX checks and 20 integration checks. These documented checks were not rerun for this page.
- Observed in use
- Documented workflows use synthetic data and include report generation and invalidation after configuration changes. They are not a study with external professionals.
- Pending
- Expand analytical cases, observe external use and review distribution terms. Software checks do not establish the validity of instruments analyzed with it.
Copies of VALIDACION_FASE3.md and UX_FASE4.md, dated September 11, 2026, are available here. Their results do not certify the validity of instruments analyzed with the software.