Pershing, Validity and Reliability in Questionnaire Development

advertisement
Questionnaire Development
Measuring Validity & Reliability
James A. Pershing, Ph.D.
Indiana University
Definition of Validity

Instrument measures what it is
intended to measure:




Appropriate
Meaningful
Useful
Enables a performance analyst or
evaluator to draw correct conclusions
Types of Validity
Face
 Content
 Criterion

Concurrent
 Predictive


Construct
Face Validity



It looks OK
Looks to measure what
it is supposed to
measure
Look at items for
appropriateness



Looks Good To Me
Client
Sample respondents
Least scientific validity
measure
Content-Related Validity

Balance
Organized review of
format and content of
instrument



Definition
Sample
Content
Format

Comprehensiveness
Adequate number of
questions per objective
No voids in content
By subject matter
experts
Criterion-Related Validity
Subject Instrument A
Task
Inventory
John
yes
Mary
no
Lee
yes
Pat
no
Jim
yes
Scott
yes
Jill
no
Instrument B
Observation
Checklist
no
no
no
no
yes
yes
yes
Usually expressed as a correlation coefficient
(0.70 or higher is generally accepted as
representing good validity)

How one measure
stacks-up against
another




Concurrent = at same
time
Predictive = now and
future
Independent sources
that measure same
phenomena
Seeking a high
correlation
Construct-Related Validity

Prediction 1 - Confirmed
T
H
E
O
R
Y
Prediction 2 - Confirmed
Prediction 3 - Confirmed
Prediction n - Confirmed


A theory exists explaining
how the concept being
measured relates to other
concepts
Look for positive or
negative correlation
Often over time and in
multiple settings
Usually expressed as a correlation
coefficient (0.70 or higher is generally
accepted as representing good validity)
Definition of Reliability


The degree to which measures obtained with
an instrument are consistent measures of
what the instrument is intended to measure
Sources of error

Random error = unpredictable error which is
primarily affected by sampling techniques



Select more representative samples
Select larger samples
Measurement error = performance of instrument
Types of Reliability
Test-Retest
 Equivalent Forms
 Internal Consistency

Split-Half Approach
 Kuder-Richardson Approach
 Cronbach Alpha Approach

Test-Retest Reliability


Administer the same instrument twice to the same
exact group after a time interval has elapsed.
Calculate a reliability coefficient (r) to indicate the
relationship between the two sets of scores.


r of+.51 to +.75 moderate to good
r over +.75 = very good to excellent
TIME
Equivalent Forms Reliability




Also called alternate or parallel forms
Instruments administered to same group at same time
Vary:
Stem:
Response Set:
-- Order
-- Order
-- Wording
-- Wording
Calculate a reliability coefficient (r) to indicate the
relationship between the two sets of scores.


r of+.51 to +.75 moderate to good
r over +.75 = very good to excellent
Internal Consistency Reliability


Split-Half
Break instrument or subparts in ½ -- like two
instruments
Correlate scores on the
two halves
Kuder-Richardson (KR)
 Treats instrument as
whole
 Compares variance of
total scores and sum
of item variances
Cronbach Alpha
Best to consult statistics book and
consultant and use computer software
to do the calculations for these tests


Like KR approach
Data scaled or ranked
Reliability and Validity
So
unreliable
as to be
invalid
Fair
reliability
and fair
validity
Fair
reliability
but invalid
Good
reliability
but invalid
Good
reliability
and good
validity
The bulls-eye in each target represents the information that is
desired. Each dot represents a separate score obtained with the
instrument. A dot in the bulls-eye indicates that the information
obtained (the score) is the information the analyst or evaluator
desires.
Comments and Questions
Download