Share

Objective Measurement of Subjective Phenomena

6. Problems in Measuring Constructs

Dealing with non-normality

Various options can be used to deal with non-normality, including:

Transforming values

If distribution is:

  • Negatively skewed, squaring values may make the distribution more normal.
  • Positively skewed, then a log transformation or square root transformation may make the distribution more normal.

Revising items to make content more (or less) extreme

If distribution is:

  • Negatively skewed, make the item harder to endorse.

Example 22

Negatively skewed example:

Consider the item “I am usually happy.” assessed on 1-to-7 scale.
  • Most persons might use scale points 5-7.
  • Few would use scale points 1-4.
Reword the item “I am the happiest person I know.”
  • Fewer persons would use the highest scale points.
  • More would use lower scale points.
  • Positively skewed, make the item easier to endorse.

Example 23

Positively skewed example:

Consider the item “I often use harsh language when disciplining my children.” again assessed on 1-7 scale.
  • Few persons would use high scale points (e.g., 4-7).
  • Most persons would use rather low scale points.
Reword the item “I occasionally use harsh language when disciplining my children.”
More persons are likely to use higher scale values, admitting to using such language only occasionally.