Measurement & Psychometrics
Measures that move. And likely mean the same thing everywhere.
Use this for
- Your current instrument is noisy, ceilinged/floored, or drifts across cohorts/devices.
- You’re adapting a scale to a new language, mode (paper→ePRO), or indication.
- You must show change (responsiveness/MCID), not just correlation.
- Reviewers/regulators will read the appendix—twice.
What you walk away with
- Measurement plan — constructs → tasks/scales → endpoints that map to decisions.
- Validation evidence — reliability, validity, responsiveness; readable stats & figures.
- Scoring rules — sum/T-scores, missingness rules, cut-points, edge-case handling.
- Traceability — item/task → endpoint table → SAP shells, ready for audits.
Patterns we reach for
- Construct first — context-of-use defines items, not tradition.
- CFA → IRT/Rasch when earned — only add complexity when relevant.
- Anchor-based MCID — distribution metrics as support, not the headline.
- Bridges — language/mode invariance and response-process checks (cognitive interviews).
Quality gates
- Reliability targets (ω/ICC ≥ 0.80) stated up front.
- Invariance/DIF documented; biased items fixed or dropped.
- Missingness budget and rescue rules defined.
- Assumptions written down — prereg where appropriate.
Rapid · 2–3 weeks
- Gap report and 3–5-option shortlist (stats + ops).
- Adoption plan: licenses, training, data flow.
- Decision memo with risks we accept (and not).
Build & Validate · 6–10 weeks
Scale development & validation
- EFA/CFA/SEM; IRT/Rasch if useful; test–retest.
- Invariance/DIF and scoring pack (code + sheet).
- Validation report + admin manual.
Analysis Pod
Blocks of expert hours
QA & preprocessing · Modeling & figures · Results write-ups.
Example runs
Instrument bake-off for a cognitive endpoint
Cross-cultural adaptation (3 languages) with DIF
Responsiveness & MCID for an app readout
Content-validity interviews (patients/clinicians)
Boundaries
- We won’t force a scale to measure what it doesn’t.
- Not all lab tasks travel to field/trial — we’ll say so.
- You (or your CRO) collect data; we design gates and do the analysis.
Turn measures into readouts.
Book a 15-minute consultation or ask for a sample validation pack.
FAQ
Do you build new scales or only adapt?
Both. If a fit-for-purpose instrument exists, we adopt; if not, we build and validate.
How big a sample do we need?
Rule-of-thumb: ≥5–10× items (≥200) for CFA; IRT/Rasch often benefits from 300–1000+, sized to your trait coverage.
Do you always use IRT/Rasch?
Only when it changes decisions (item precision, thresholds, adaptive forms). Otherwise, lean and clear wins.
How do you set MCID?
Anchor-based first (e.g., PGI/CGI-I/known-groups), with distribution methods as support and precision (CIs) reported.
Can this plug into a CNS trial?
Yes — endpoint table, visit timing, and SAP shells included on request.
Need Some Help?
Feel free to contact us for any inquiry or book a free consultation.
