Home
| Statistics
and
Research Methods | Psychometrics | Human-Computer
Interaction
| Organisational
Psychology | Psychology
for Work
Psychometrics
Hints
and Tips
SO,
if you've arrived at this page, then I assume you need some help with
deciphering statistical information. Bear in mind that the information
below is a rough guide only, and that there are many nuances involved
in Statistics which really can't be summarised in a few sentences!
What kind of Statistics should be done, and what do they mean?
Item
analysis -this is an analysis of the items, or questions in the
test. This involves several steps but should tell the manufacturer
which items are working hard, which items are repetitions, and whether
the items relate together as predicted (e.g. all the items relating to
Extraversion do in fact relate together, and crucially, do not relate
to items not relating to Extraversion).
Construct
analysis - this relates to the constructs, or groups of items
in the test (e.g. Extraversion is a construct). This analysis should
tell you that the constructs relate together in the way predicted (e.g.
that Introversion and Extraversion can be viewed as opposite ends of a
continuum)
Reliability
Analysis - this analysis should be conducted using a large group
of people who have taken the test, and (using a variety of techniques)
their results should be compared to assess test performance
All of the above analysis involves Statistical techniques such as
Principle Components Analysis (Factor Analysis) and Confirmatory Factor
Analysis
Coefficient Alpha
Many text books suggest that coefficient alpha is the definitive
statistic which will tell you all you need to know about the test in
question. This statistic is indicative of the reliability, or internal
consistency of the test, but beware! This statistic can be misleading
if there is insufficient distinction between constructs (i.e.
everything relates to everything else).
If you need more help than this, please contact Jane at one of the
email addresses below.
Back