Score Linking and Equating

Description

Makes scores from two questionnaire versions or instruments comparable by matching percentiles or means and SDs, using a data-collection design that controls for group differences.

When a questionnaire is reworded, translated, shortened or replaced, old and new scores are not automatically comparable. Linking puts them on a common scale; equating is the strict case where both forms measure the same construct with equal reliability, so scores become interchangeable. Three decisions are involved: defining comparable scores (usually equal percentile ranks in a target population), choosing a data-collection design, and estimating the linking function. Equipercentile linking gives a non-linear conversion when the score distributions differ in shape; if they differ only in mean and SD it reduces to a linear function. Designs are single group (everyone completes both forms, counterbalanced), equivalent groups (random or spiralled assignment of forms), and anchor-test designs with common items, which adjust for group differences when randomisation is impossible. Linking two different instruments, such as two satisfaction scales, gives a concordance table rather than true equating.

Attributes

Loss type
Unmigrated
Effort
Weeks

Steps

  1. Define comparable scores, usually equal percentile ranks in the target population of users.
  2. Choose a design: single group with counterbalanced order, equivalent groups via random assignment of forms, or an anchor design with shared items across forms.
  3. Collect data on both forms (or form plus anchor) from samples large enough for stable percentiles, typically several hundred per form.
  4. Estimate the linking function: linear (matching means and SDs) if distributions have similar shape, otherwise equipercentile with smoothing.
  5. Apply the conversion, then check it on a hold-out sample and inspect the standard error of the conversion across the score range.
  6. Document whether the result is true equating or only a concordance, and limit claims accordingly.

Sources

  • The SAGE Handbook of Quantitative Methods in Psychology (Millsap & Maydeu-Olivares, 2009)

Also known as

  • Test equating
  • Linear linking
  • Equipercentile equating
  • Anchor-test design
  • Score concordance between instruments

Categories

  • Analytics & Behaviour
  • Survey & Psychometrics