| |
In order for a personality assessment survey be acceptable under the standards and guidelines
of the American Psychological Association, the U.S. Equal Employment Opportunity Commission (EEOC),
the Canadian Psychological Association, and the British Psychological Society various steps have to be taken.
First, the items, whether they are phrases, words, questions, pictures, etc., have to measure what they
purport to measure - they must be valid. For example, is choosing the word "self-confident" to describe
oneself actually a measure of the personality trait ego-centeredness? Does someone who prefers to go to
a party rather than read a book more of an extrovert than an introvert? Men pick red as their favorite
color more often than woman. Does this mean something? Test developers have to use a statistical technique
called factor analysis to sort and label items. Also, it is during this stage the researcher determines if
any of the items have bias. In other words, is an item more attractive or unattractive to a specific group
- gender, race, ethnicity? Biased items should be eliminated or accounted for in test construction. Assuming
that an item measures a particular trait (construct) or is unbiased is insufficient. Publishing factor analysis
results does not compromise trade secrets and reputable test developers publish their findings and don't hide
behind a veil of "proprietary information".
Second, the items, both individually and collectively, have to be reliable. That is, the items are stable and
are not subject to whim, mood, or other easy influences. For example, the word "happy" in describing oneself
says nothing about the trait makeup of an individual. He or she may be happy this morning and, due to a minor
incident, be unhappy in the afternoon. A multiple-choice arithmetic test given to a class of 6th graders generates
an average test score of 80% on Monday. On Friday the same test generated an average score of 60%. Obviously, the
test is not measuring actual student knowledge; they are guessing. Again, reliability coefficients (.70 is the
generally accepted cut-off number) must be published.
Translations of tests and surveys have to be separately validated. Words and phrases can take on different
meanings and norms or average scores can differ.
Third, the items or, once grouped, constructs must have a practical value; they must be job relevant.
The Supreme Court of the United States has repeatedly affirmed the legality and value of "testing"
starting in 1971 with Griggs v. Duke Power Co. In that decision the Court stated that as long as
test results correlated with job performance they were not only permissible but also "desirable".
A test or survey that measures irrelevancies or which are prohibited by law (e.g. "What race are you?
Do you plan on having children?") clearly should not be used.
OAD conducts job validity studies for its clients; statistically correlating traits and job behaviors with relevant performance measures.
For additional information, please contact Michael Gray at mgray@oadllc.com or 781-878-6387.
|
|
