Chapter 3

Basic measurement and scaling

concepts
The map for the journey
 Connecting
List all the things that you could use to
measure the size of this lecture room, as
a new carpet needs to be ordered for it.
What criteria should the measurement
method meet for you to be sure that the
correct-sized carpet will be ordered?
Do all the measurement methods on
your list meet these criteria?
 Mapping the chapter
• Defining measurement and
psychological measurement
• Properties of measurement scales
• Categories of measurement levels
• Measurement error
• Measurement scales
• Basic statistical concepts
• Norms
• Setting standards and cut-off scores
 Mapping the chapter
• Measurement • Measurement
• Psychological levels – properties
measurement & categories
• Measurement
errors
• Measurement
Measureme Scales – scaling
options &
nt levels, scale format
Definition
errors, and
scale

Meaningful •
interpretati Statistical
on concepts • Displaying data
• Norms • Central tendency
• Standards • Variability
• Cut-scores • Association
 Definition of measurement
‘Measurement is the assignment of
numbers to objects or events according to
sets of predetermined (or arbitrary) rules,
or to frame it more precisely in
psychometric terms, the transformation of
psychological attributes into numbers’
(Runyon & Haber, 1980, p. 21)
Levels of measurement
 Three properties of
measurement scales
• Magnitude – the property of ‘moreness’.
• Equal intervals – the differences between all
points on the scale are uniform and continuous
data is generated. Where the difference between
points on a scale is not the same, categorical data
is generated.
• Absolute zero – there is nothing present
of the attribute being measured. With many
human attributes it is difficult, if
not impossible, to define an absolute
zero point.
Categories of measurement levels

• Measurement serves different functions,

such as sorting, ordering, rating, and
comparisons.
• Numbers are used in different ways to:
– name (i.e. group according to a label ̶
nominal numbers)
– represent a position in a series
(i.e. to order ̶ ordinal numbers)
– represent quantity (i.e. rating or comparing ̶
cardinal numbers)
 Categorical data
• Nominal – numbers are assigned to an
attribute to describe or name it
E.g. official language coded from 1 to 11
• Ordinal – numbers are assigned to objects
that reflect some sequential ordering or
amounts of an attribute
E.g. educational qualification coded from
1 (primary school) to 5 (postgraduate degree) =
5 is ranked higher than 4, which is ranked higher
than 3, etc.
 Continuous data
• Interval – equal numerical differences can
be interpreted as corresponding to equal
differences in the characteristic being
measured
[….insert example here X]
• Ratio – equal numerical difference can be
interpreted as reflecting equal differences
in the characteristic being measured, AND
there is a true zero
Measurement errors
Two types of measurement errors
• Random measurement error
– ‘Variability of errors of measurement that function in a
random or nonsystematic manner’ (Price, 2017, p. 254).
– Arise from chance factors such as guessing, shifts in
attention, disruptions during assessments.
• Systematic measurement error
– ‘Constant errors of measurement that occur
when all test scores are excessively high or low’ (Price,
2017).
– When test conditions are generally better for one group
over another (e.g. good lighting versus poor lighting;
noisy venue versus a quiet venue, etc.).
Measurement scales
 Measurement scales
• Category scales
• Likert-type scales
• Semantic differential scales
• Intensity scales
• Constant-sum scales
• Paired-comparison scales
• Graphic rating scales
• Forced-choice scales
• Ipsative scales
• Guttman scales
 Category scale
(Figure 3.1)
How frequently do you make use of public
transport?

All the time

Sometimes
Rarely
Never

Often
(1) (2) (3) (4) (5)
 Likert-type scale
(Figure 3.2)
My supervisor can be described as an
approachable person.
Strongly disagree

Strongly agree
Uncertain
Disagree

Agree
(1) (2) (3) (4) (5)
 Semantic differential scale
(Figure 3.3)
How would you best describe yourself?

Emotional 1 2 3 4 5 6 7 Intellectual

Conventional 1 2 3 4 5 6 7 Modern

Lazy 1 2 3 4 5 6 7 Active

Traditional 1 2 3 4 5 6 7 Progressive
 Intensity scale
(Figure 3.4)

How frequently are you in disagreement

with your supervisor?

Never 1 2 3 4 5 6 7 Always
 Constant-sum scale
(Figure 3.5)
Divide 100 points among the following
characteristics of customer service you
have received from company A:

Delivery on time 30
Quality products 50
Affordable price 20
100 points
 Paired-comparison scale
(Figure 3.6)
Divide 100 points between the two
opposing values

Professionalism 20 80 Integrity

Integrity 60 40 Transparency

Competence 55 45 Professionalism
 Graphic rating scale
(Figure 3.7)

How happy do you feel today?

Very unhappy Very happy

 Forced-choice scale
(Figure 3.9)
How do you typically respond in a conflict
situation? (Tick the box that describes you
best)
I dislike conflict. I avoid it
whenever I can
I feel neutral about conflict. I do
not search or avoid it
I thrive on conflict. I will search for
conflict situations
 Ipsative scales
(Figure 3.10)
Place an ‘M’ in the box that is most like you,
and an ‘L’ in the box that is least like you

I dislike conflict. I avoid it

whenever I can
I feel neutral about conflict. I do
not search or avoid it.
I thrive on conflict. I will search for
conflict situations
 Guttman scale
(Figure 3.11)
My attitude about crime is based on my
experience of an incident that happened in …
(Circle ‘Yes’ or ‘No’ for each of the following:)

Yes No
The vicinity of the neighbourhood I live in 1 0
The neighbourhood I live in 1 0
The street I live in 1 0
The house of a next-door neighbour 1 0
My own home 1 0
 Designing a measurement scale:
Considerations
• Will the scale measure one dimension or many
(multi) dimensions or facets of a construct?
• Will the item be formulated as a question or
a statement?
• Will all the response labels have descriptors or
only the extreme categories?
• Will the scale measure one construct or
many constructs?
• How many response categories are needed?
• Will the scale lead to ipsative or normative
comparisons?
 Basic statistical concepts
• Data can be depicted graphically to make sense of
it.
• The centre of the distribution can be determined
by calculating the mean, median, and mode.
• The distribution of the scores around the mean
can be computed (range and standard deviation).
• Two sets of scores can be correlated and
performance on the one can be predicted from
the other.
Displaying data

Normal distribution
(bell curve)

Negatively skewed

Positively skewed
 PRACTICE EXAMPLE
Five potential trainee managers are assessed on a Learning Potential
Test to identify those who could attend a management development
programme (minimum test score = 0; maximum = 5). Their
performance is as follows:
3.5

3
3
2.5
Trainee 1
2 Trainee 2
2 2
1.8 Trainee 3
1.5
Trainee 4
1 Trainee 5
1
0.5

0
Learning Poterntial Test Performance
Measures of central tendency
MEAN MEDIAN MODE

Presents the score where,

Sum of scores divided when the data are
by number of scores arranged in numerical The most
order, 50% of the frequently
distribution is below that occurring score
point and 50% above it;
the 50th percentile

1.96 2
2
 Measures of variability
STANDARD
RANGE VARIANCE
DEVIATION

Indicates how much

The difference How scores vary
the scores in a
between the largest about the mean or
population of test-
and smallest score how they deviate
takers should be
from the mean
spread out (bunched
up or scattered)

1 0.51 0.71
 Measures of association
CORRELATION REGRESSION

• Is one variable related to

another variable? Can one variable be
• If one, then the other predicted from another
• The variables appear variable?
together
 What is a norm?
A norm can be defined as a measurement
against which an individual’s raw score is
evaluated so that the individual’s position
relative to that of the normative sample
can be determined.
 Establishing norms
• Choose a norm group and identify the
subgroups from the norm group that
must be represented.
• Administer the test to a representative
sample from the norm group.
• Standardised or normal scores are then
calculated.
• Test-taker’s score is then compared to
the standardised or normal scores.
 Co-norming
Co-norming entails the process where
two or more related, but different
measures are administered and
standardised as a unit on the same
norm group.
 Types of norms
• Developmental scales

Mental-age scales Grade equivalents

• Percentiles
• Standard scores (Z-scores, normalised
standard scores such as the T-score,
stanine scale, and sten scale)
• The deviation IQ scale
 Standards and cut-off scores
• Instead of finding out how an individual has performed
in relation to others, you can compare performance to
an external criterion (standard).

• For example, in the psychometrics examination that is

set by the Professional Board, the passing standard has
been set at 70 per cent. Someone who obtains a mark
of 60 per cent may have performed better than a
classmate who obtained 50 per cent (norm-referenced
comparison), but would fail the examination, as his or
her percentage is below that of the 70 per cent
standard (cut-off) that has been set.
 Expectancy table
APTITUDE NUMBER JOB RATING: JOB RATING:
SCORE OF SATISFACTORY UNSATISFACTORY
PEOPLE (PERCENTAGE) (PERCENTAGE)
16–20 10 90 10

11–15 8 80 20

6–10 15 30 70

Below 5 10 10 90
 Cut-off scores (cont.)
• Main advantage is that they offer an easily
understandable way of interpreting the relationship
between test score and probable levels of success on a
criterion.
• Large sample sizes are required as correlation data can
be unstable from sample to sample.
• A criterion can change over time and impact on the
accuracy of decisions.
• Rather use a band of cut-off scores than a single one
• A measure may have differential predictive validity for
various subgroups (e.g. race or gender) – a
cut-off score for the whole group could lead to
bias against subgroups.
Ethical dilemma case study 3.1
Dr Phumzile Mthembu is a specialist consultant
in psychological assessment and selection.
Dr Mthembu was requested by a mining
company to select a large number (>200) of
management trainees from different culture
groups to attend an accelerated management
development programme.
Dr Mthembu used a measure to test applicants’
learning potential and found that there were
significant differences in the culture groups’
mean scores and standard deviations.
Ethical dilemma case study 3.1
• Should Dr Mthembu reveal this to the
company or should he just make statistical
corrections on the respective groups’ test
scores and proceed with the selection?
• Who are the potential stakeholders that
need to be informed about the test’s
failure to show similar central tendency
and dispersion between the respective
cultural groups?