Test Construction: Creating Parallel Forms

1
Test Construction Creating Parallel Forms

• Oleksandr Chernyshenko and Stephen Stark
• University of Illinois at Urbana-Champaign

2
Test Development Using Classical Test Theory

• Items with high item-total correlations are
  retained to form a scale with high internal
  consistency
• Distribution of total test scores is obtained and
  compared to the distribution desired by the test
  developer
• Some item switching is done to obtain the closest
  match between desired and obtained total score
  distributions.
• Parallel forms are created to have identical test
  score distributions.
• Equivalence of test score means, variances, and
  errors is interpreted as evidence that tests are
  parallel.

3
Limitations of Classical Test Theory Approach

• Statistics are sample dependent
• P-values and item-total correlations can vary
  across samples
• Depend on set of items examined
• Standard Error of Measurement is an average
  across trait levels
• No information regarding errors at specific score
  levels

4
Advantages of IRT

• Item parameters are sample invariant
• No assumptions about underlying distribution of
  theta
• Item parameters do not depend on subset of items
  examined
• Can examine contributions of individual items to
  test characteristics
• Obtain Conditional Standard Error of Measurement
• Compute SE at specific trait levels using test
  information

5
IRT Functions Used for Test Construction
6
IRT Functions Used for Test Construction
TCC
7
ConstructingTests to Measure Wide Range of Skill
8
ConstructingTests for Targeted Ability Range
9
ConstructingParallel Tests