Concurrent Validity: Equivalence Testing and Model Two Regression

Equivalence Testing

  • A statistical test used to assess agreement, becoming increasingly popular.

  • Commonly used in:

    • Pharmacology: FDA approval of medical treatments and drugs.
    • Nutrition: Assessing if a supplementation or dietary protocol is equivalent to a placebo effect.
    • Exercise Science: Evaluating if wearables accurately estimate energy expenditure.

  • Example: Comparing maximal running velocity from a GPS unit to a radar gun.

  • Built around two one-sided t-tests.

  • If the 90% confidence interval of the mean difference falls within an a priori decided acceptable difference, the measures are considered equivalent.

  • Null Hypothesis:

    • Traditional null hypothesis testing: No difference in population means.
    • Equivalence testing: The means of measures are not equivalent (they are different).

  • Equivalence testing examines only the mean difference, similar to traditional null hypothesis testing.

  • Random error is important and can be assessed using limits of agreement.

  • Recommendation: Use limits of agreement and equivalence testing together.

    • Formal statistical test for equivalence.
    • Clinical judgment for agreement.

Model Two Regression

  • A different form of regression than ordinary least squares.

  • Used for assessing agreement.

  • Various forms exist, including:

    • Ordinary least products (most popular).
    • Weighted least products (a variant of ordinary least products).