You are on page 1of 44

Quality of Measures

• Definitions
– Correlation, Reliability, Validity, Measurement
error
• Theories of Reliability
• Types of Reliability
– Standard Error of Measurement
• Types of Validity
• Article
• Exercise
Definitions

• Correlation
– reflect direction (+/-) & strength (0 to 1) of the
relation between two variables

• Variance explained
– Reflects the strength of relation of two variables
• Square of correlation
• Varies from 0 to 1
250

Vince Carter
210
Weight (pounds)

170 Tom Cruise

130

Julia Roberts
Calista Flockhart
90
150 160 170 180 190 200
Height (cm)
250
r = .76
Vince Carter
r2 = 58%
210
Weight (pounds)

170 Tom Cruise

130

Julia Roberts
Calista Flockhart
90
150 160 170 180 190 200
Height (cm)
Effect of Measurement Error on
Correlations
200
r = 1.00
190 r2 = 100%

180
Height (cm)

170

160

150
150 160 170 180 190 200
Height (cm)
200

r = .98
190
r2 = 96%
Self-Reported Height (cm)

180

170

160

150
150 160 170 180 190 200
Objective Height (cm)
250
r = .92; r2 = 85%

225
Self-Reported Weight (cm)

200

175

150

125

100
100 125 150 175 200 225 250
Objective Weight (cm)
Definitions
• Reliability
• Consistency & stability of measurement
• Reliability is necessary but not sufficient for
validity
• E.g. A measuring tape to is not a valid way to measure
weight although the tape reliably measures height and
height correlates w/weight
• Validity
• Accuracy/meaning of measurement

• Example: unstructured vs. structured job


interviews
Theories of Reliability

• Classical Test Theory explains random


variation in a person’s scores on a measure
• Effects of learning, mood, changes in
understanding etc.
• Test score=true score + error
• Errors have zero mean
• Errors are uncorrelated with each other
• Errors are uncorrelated with true score
• Constant error is part of true score
Types of Reliability

• Test-retest
• Consistency across time
• Parallel forms
• Consistency across versions
• Internal
• Consistency across items
• Scorer (inter-rater)
• Consistency across raters/judges
Example: The Satisfaction with Life Scale
(SWLS)
1. In most ways my life is close to ideal.
2. The conditions of my life are excellent.
3. I am satisfied with my life.
4. So far I have gotten the important things I
want in my life.
5. If I could live my life over, I would change
almost nothing.
1 2 3 4 5 6 7
Strongly Strongly
Disagree Agree
Types of Reliability

• Test-retest reliability
• Correlation of scores on the same measure taken at
two different times
• Time interval assumes no memory/learning effects

• Parallel-forms
• Correlation of scores on similar versions of the
measure
• Forms equivalent on mean, stan dev, inter-correlations
• Can have time interval b/w admin of two forms
Types of Reliability
Test-retest Reliability
Time 1 Time 2
I1 I2 I3 AvgT1 I1 I2 I3 AvgT2
P1
P2
P3
Correlate AvgT1 to AvgT2 to get reliability

Parallel Forms P=participant


Version 1 Version 2 I=item
I1 I2 I3 AvgV1 I1 I2 I3 AvgV2
P1
P2
P3
Correlate AvgV1 to AvgV2
7
r = .73; r2 = 50%

6
SWLS Time 2 (End of Semester)

1
1 2 3 4 5 6 7
SWLS Time 1 (Beginning of Semester)
Test-retest reliability of SWLS
• Good test-retest reliability
•Participants have similar scores at Time 1
(beginning of semester) and at Time 2
(end of semester).
•Retest reliability is useful for constructs
assumed to be stable
•Current mood (e.g., how you feel right
now) shows low-retest correlations, but
that does not mean that the mood measure
is not reliable
Types of Reliability

• Internal Consistency
• Correlation of scores on two halves of the measure
• Length of measure increases reliability

• Inter-rater
• Correlation of raters’ scores
• E.g., Scores on structured job interview
• Can also include time interval
– e.g., ratings of the worth of jobs across time & across judges
Types of Reliability
Internal Reliability
Half 1 Half 2
I1 I2 I3 AvgH1 I4 I5 I6 AvgH2
P1
P2
P3
Correlate AvgH1 to AvgH2

Inter-rater Reliability
Rater 1 Rater 2
I1 I2 I3 AvgR1 I1 I2 I3 AvgR2
P1
P2
P3
Correlate AvgR1 to AvgR2
7
r = .70; r2 = 49%

6
SWLS Items 3, 4, & 5

1
1 2 3 4 5 6 7
SWLS Items 1 & 2
Internal consistency of SWLS
• Satisfactory internal consistency.
•Participants respond similarly to items
that are supposed to measure the same
variable.
•Should be .70 or higher

•Measurement error accounts for half of the


variance in SWLS scores.
Types of Reliability

• Test-retest
• Parallel forms
• Internal
• Scorer (inter-rater)
Standard Error of Measurement

• SD of scores when a measure is completed


several times by the same individual
• Mostly used in selection contexts
• Decide which of two individuals are hired
• Decide whether a test score is significantly higher/lower
than a cutoff score
Correction for Attenuation

• Real correlation between two variables after


removing unreliability of each measure
• Divide observed correlation by product of the
square roots of individual reliabilities
• Note: Selection research only controls for unreliability
in criterion bec. we are more interested in the value of
the predictor given a perfectly reliable criterion
Quality of Measures

• Definitions
– Correlation, Reliability, Validity, Measurement
error
• Theories of Reliability
• Types of Reliability
• Standard Error of Measurement
• Types of Validity
Validity
Evidence that a measure assesses the construct
Reasons for Invalid Measures
• Different understanding of items
• Different use of the scale (Response Styles)
• Intentionally presenting false information
(socially desirable responding, other-
deception)
• Unintentionally presenting false
information (self-deception)
Types of Validity

Criterion Content Construct


Validity Validity Validity

Predictive Concurrent Convergent Discriminant


Validity Validity Validity Validity

Adapted from Sekaran, 2004


Types of Validity

• Content Validity
• Extent to which items on the measure are a good
representation of the construct
• e.g., Is your job interview based on what is required for
the job?
• Content validity ratio based on judges’
assessments of a measure’s content
• e.g., Expert (supervisors, incumbents) rating of job
relevance of interview questions
Types of Validity

• Criterion-related Validity
• Extent to which a new measure relates to another
known measure
• Validity coefficient= Size of relation between the new
measure (predictor) and the known measure (criterion)
(a.k.a correlation)
• e.g., do scores on your job interview predict
performance evaluation scores?
Types of Criterion Validity
• Concurrent
• Scores on predictor and criterion are collected
simultaneously (e.g., police officer study)
• Distinguishes between participants in sample who
are already known to be different from each other
• Weaknesses
• Range restriction
– Does not include those who were not hired, fired & promoted
• Differences in test-taking motivation (employees vs.
applicants)
• Experience with job can affect scores on criterion
Types of Criterion Validity
• Predictive
• Scores on predictor (e.g., selection test) collected
some time before scores on criterion (e.g., job
performance)
• Able to differentiate individuals on a criterion
assessed in the future
• Weaknesses
• Due to management pressures, applicants can be chosen
based on scores on predictor (can have range restriction,
but this can be corrected)
• Often, special measures of job performance are
developed for validation study
Correction for range restriction

• When full range of scores on predictor


variable is available
– Use unrestricted and restricted standard
deviations of predictor variable & the observed
correlations b/w predictor & criterion
Types of Validity (cont’d)

• Construct Validity
• Extent to which hypotheses about construct are
supported by data
1. Define construct, generate hypotheses about
construct’s relation to other constructs
2. Develop comprehensive measure of construct & assess
its reliability
3. Examine relationship of measure of construct to other,
similar and dissimilar constructs

• Examples: height & weight; Learning Style


Orientation measure; networking; career outcomes
Establishing Construct Validity

• Multi-trait multi-method matrix


• Convergent validity coefficient
• Absolute size of correlation between different measures
of the same construct
• should be large, significantly diff from zero,
• Discriminant validity coefficient
• Relative size of correlations between the same construct
measured by different methods compared to
• Different constructs measured by different methods
• Different constructs measured by same method (method bias)
Corr b/w Objective (O) & Self-
Reports (SR) of Height & Weight

O-H SR-H O-W SR-W

O-H 1.00

SR-H .98 1.00

O-W .55 .56 1.00

SR-W .68 .69 .92 1.00


Establishing Construct Validity

• Multi-trait multi-method matrix


– Different measures of the same construct should be
more highly correlated than different measures of
different constructs
• e.g., Perceived career success & promotion vs.
networking vs. promotion/salary
– Different measures of different constructs should
have lowest correlations
• e.g., Networking vs. promotion/salary
Learning Style Orientation Measure
• Item Development Study (generate
critical incidents)
– N=67
– Yes/no responses to statements
– Recall of learning events
• Two types of learning: theoretical, practical
• Two types of outcomes=success, failure
• 2 x 2 events per participant
• 112 items constructed in total
Learning Style Orientation Measure
• Item Development Study (questionnaire)
– N=154
– 112 items, 5 point likert scale (agree/disagree)
• 5 factor solution w/factor analyses
• 54 items
• Content validity sorting by 8 grad students
– Goldberg personality scale
Learning Style Orientation Measure
• Item Development Study
• Correlations b/w LSO & personality
• Only 1 sig correlation b/w 5 factors of LSOM!
• High reliabilities of subscales of LSOM (.81-.91)
• Construct (not really convergent) validity
– r b/w LSOM & personality subscales
• .42 to -.26.
Learning Style Orientation Measure
• Validation Study
– N=350 -193
– LSOM, Personality, old LSI, preferences for
instructional & assessment methods
• Construct validity
– r b/w LSOM subscales & old LSI= .01 to .31
– r b/w LSOM & personality subscales= .01 to .55
– Confirmatory factor analysis
• 5-dimensions confirmed
• High reliability
Learning Style Orientation Measure

• Validation Study
– Incremental validity
• Additional variance explained (LSOM vs LSI)

DV LSOM LSI
Subjective assessment .15 .01
Interactional instruction .21 .04
Informational instruction .06 .00
In-class Exercise
• Brainstorm constructs to develop measures
• E.g. Dimensions of CIR professor effectiveness, CIR
student effectiveness
• Choose two constructs that can be measured
similarly and be defined clearly
• Example measures
– Self-report (rating scales)
– Peer/informant reports
– Observation
– Archival measures
– Trace measures etc etc.
In-class Exercise
• Form two-person groups to
• Generate items of the 2 different measures for each of
the two constructs
• Appointed person collects all items for both
measures for both constructs
• Compiles & distributes measures to class
• Class gathers data on both measures & both
constructs
• Class enters data into SPSS format
• Compute reliabilities,means, correlations
Fill in the correlations

C2 C1
M1
M1

M2
M2
Types of Validity

Criterion Content Construct


Validity Validity Validity

Predictive Concurrent Convergent Discriminant


Validity Validity Validity Validity

Adapted from Sekaran, 2004

You might also like