Tuesday, January 25, 2011

Modern Measurement: Dimensional Scaling and Scoring : Chapter 6



Defining Scaling and Establishing its Need
-          Score scale
-          Scaling defined
o   Numerically ordering objects of an assessment, or more precisely, attributes of the object.
o   Orders the variable
o   Eg
§  Male / female – unusual in that it is not hierarchical
o   Ordinal
o   Most scales are interval
o   Employing the proper scale
§  Eg
·         Likert – 7 points
·         IQ  - 40 – 160
§  Audience
o   Indifference in the scale
§  Scale must be indifferent to object being measured.
§  Eg – gender of examinee
o   Need for scaling
§  Interpretable / useful

Reliably Making Judgments

-          The Just Noticeable Difference
o   We can tell differences
o   Threshold – point along continuum where we can agree a change has occurred
-          Fechner's Law

Early to Current Scaling Models

-          Deterministic and Probability Scaling Models
o   Probabilistic models estimate likelihood of an event, such as chance of observing a certain test score given a particular set of conditions (ie item difficulty, trait level of examinee)

Dimensional Scaling

-          Dimensionality Explained
o   Unidimensional scaling
§  Test is appraising one only construct / trait
o   Multidimensional scaling
§  Eg
·         Extraversion and self-efficiacy are distinct phenomena, and each can be hypothesized as a latent construct. When considered concurrently, however, their individual variances overlap, meaning that their latencies have an oblique relationship.
·         Multidimensional models appraise distances between traits and the specified dimensions
·         ….  When traits have a lot in common ….


Types of Scales

-          General conditions for ordered scales
-          Levels of measurement
o   Nominal
o   Ordinal
o   Interval
o   Ratio

Intervals For Scales

-          Equal and nonequal interval scales
-          Non-equal interval scale conveys more info re distribution than equal interval scales


Types of Measurement

-          Ipsative vs normative measurement
-          Ipsative – raw score that sums to constant for a given examinee
o   Forced choice formats
-          Isomorphic measurement
o   Direct one-to-one relationship exists between entities used for measurement and the object that each represents.
o   Most nominal level measurement
o   Isomorphic transformation à recoding

Derived Scales

Eg
-          Standard score
-          Percentiles
-          Scaled scores
-          Normal curve equivalents
-          Grade level scores
-          Stanines

Basic Transformations in Score Scales

Non-linear transformations

Transformations for Development Scales

Characteristics of score scales







GW Questions / Comments

-          Multidimensional tests : is it the case of main effects and interaction effect.

No comments: